In this episode of the Restaurant Technology Guys podcast, host Jeremy Julian interviews Maria, co-founder of Palona AI. They discuss how Palona is transforming the restaurant industry through generative AI, voice automation, and multimodal technologies to enhance customer experience and operational efficiency. With a background in leading tech companies like Google and Meta, Maria explains the capabilities of AI in hyper-personalization, conversational marketing, and overcoming challenges such as AI hallucinations. The episode also covers the onboarding process for AI solutions in restaurants and how these technologies can provide consistent and individualized guest experiences.
00:00 Palona.Ai
01:03 Introduction and Guest Welcome
01:40 Guest Background and Experience
03:04 Defining AI and Its Impact
07:04 AI in the Restaurant Industry
08:47 Multimodal AI Solutions
11:14 Human-Like AI Interactions
17:04 Standardizing Guest Experience with AI
20:05 AI-Powered Customer Interactions
21:04 Personalized Dining Experiences
22:34 Implementing AI in Restaurants
25:11 Guest Identification and Data Management
26:04 Ensuring AI Reliability and Security
28:48 Handling Special Requests and Continuous Learning
32:31 Conclusion and Final Thoughts
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Jeremy Julian (00:01):
In today's episode, we are joined by Maria,
who is uh, one of the foundersof palona.
Pal's really solving the world'sproblems as it relates to voice
automation, and not justordering, but all of the
different items that you can usevoice for.
To deliver a better guestexperience.
(00:22):
She's got a fantastic pedigreewhere she's worked at a lot of
large brands and is using all ofthat skill and the relationships
that she has and the team thatshe has built to really change
the world of restaurantautomation and how it engages
with voice Sta until the end,where we talk a lot about how
restaurants can implement thissolution today to ultimately
(00:46):
drive bottom line profit totheir brand.
If you don't know me.
My name is Jeremy Julian.
I am the Chief Revenue Officerfor CB S North Star.
We wrote the North Star point ofsale solution for multi-units.
Check us out@cbsnorthstar.comand now onto the episode.
I.
Welcome back to the RestaurantTechnology Guys podcast.
I thank everyone out there forjoining us.
(01:07):
Like I say, each and every time,I know you guys got lots of
choices.
So thanks for hanging out.
Today is, um, we're gonna talk alot about kind of AI and really,
quite frankly, with our gueststalk a lot about.
How the consumer continues tochange and how some of the
non-restaurant tech is really,um, making a huge impact in our
(01:28):
space.
And so I'm gonna introduce youto our guests.
Maria, why don't you introduceyourself.
Let's talk a little bit aboutwhere your background is and
where you came from, and then wecan talk about what you've built
and what your latest project is.
Maria Zhang (01:40):
Yeah.
Hi Jeremy.
Thank you so much for having meon the show.
I.
From you and share a little bitfrom my story.
Yeah.
So I stu I studied computerscience and have been in this
industry, for over two decades.
Um, most recently I was VP ofengineering at Google.
I supported a large global,product development team of,
(02:00):
several hundred people.
and before that I was at meta.
And, I also have done a startup,back in, um, thousand 10.
I left Microsoft and when iPhonecame out, I got really excited
and the product I built wascalled Alike.
It's a local discovery andrecommendation tool.
So during that time, already getto work with, local restaurants
(02:22):
and such, and how to make smartrecommendations.
And that product was, acquiredby Yahoo in 2013.
So fast forward, this companyPalona.
I started with a fewco-founders, last year in 2024.
and we really excited about thisgenerative AI technology.
And then we want to bring themost advanced, multimodal,
(02:45):
multimodality generative AIsolution to serve, restaurants
and make the guest experiencebetter and make the operations,
more seamless and more costeffective.
Jeremy Julian (02:55):
Wow, you got your, you got your pitch deck
down.
I like it.
Um, as far as where it came fromand the fact that you've got
such experience, Microsoft Meta,um, Google is awesome.
I.
We were only, about 55 secondsin before AI came up.
And so Maria, I would love foryou to define what is ai,
because I think there's a lot ofpeople that it's a misnomer and
(03:16):
being in the tech space, you'vebeen in the tech space for a
long time.
There's a lot of these thingsthat have been happening for
some time now.
But I think because of.
Open AI and chat.
GPT has really put it inconsumer's hands in a way that,
I think, even five years ago orthree years ago, nobody really
knew while some of this stuffwas happening in the background
(03:36):
with, with Google and some ofthese other large companies, now
it's become almost mainstreamwhere you talk about it.
And so define what it is andwhat is this large language
model and why do you think ithas such an opportunity to solve
so many problems within therestaurant space?
Maria Zhang (03:50):
Yeah, that's a great question.
Yeah.
So, you are absolutely right.
the term right now is bothubiquitous and honestly a little
overused.
Um, but AI is not a new concept,right?
this concept started decadesago.
and I think as a.
Say 20 years ago, we avoid, usethis term artificial
(04:11):
intelligence because they soundso dystopic at all the movies we
all watch the Sky night andsuch.
And so, so we, use alternativeterms like machine learning,
predictive models.
lemme give you example.
I worked at Zillow in know veryearly days of Zi.
There's.
Long, long time ago, 2006, Ithink.
but if you think about Zillow,that is definitively artificial
(04:33):
intelligence because you could,spend what,$1,200 and hire
assessor to give you a estimatorassessment value.
Or you can go to Zillow and havethe algorithm right.
Gave you estimate, which is.
All, but then Zillow positionedthemselves as useful product,
right?
Um, so fast forward, I think,was 2022 a chat.
(04:58):
GPT came out in 2324, as you cansee, spearheaded by AI and
Microsoft copilot, Google Jamand Perplexity.
the name just layers on AppleIntelligence, right?
Amazon has a.
Is it's inflection point.
So a couple things happened.
one is, they're called largelanguage models.
and it's a descrip.
(05:18):
They're very large.
They have, many billions of,perimeters, parameters.
And what made this.
possible is, high performancecomputing.
And that's why Nvidia stock isso high because of GPU.
And then there was anotheralgorithm called Transformer
that really empowered, massiveparallel computing.
(05:39):
So this is another, Of computerscience called high performance
computing.
So it's h PC that empowered verylarge models that enabled all
these capabilities.
Yeah, so, so they're generative,right?
Meaning they can, not onlyreturn a paragraph, but even a
article even, you can write anovel if you want, if you wanna
read that novel, a large amountof text, This is I mentioned
(06:03):
earlier is multimodality.
Jeremy Julian (06:06):
and I think even going back to some of your early
days, it's amazing becauseagain, it's been around, we
think about the friendsuggestions that are on your
Facebook feed or on your Twitterfeed, or just even what shows up
in your feed.
When I sit next to my wife onthe couch and we're both
scrolling, Instagram, herInstagram feed looks very
different than mine when I loginto Gmail.
Um, her Gmail looks verydifferent than mine, and I
(06:28):
think.
Because the world of compute hasbecome so ubiquitous, as you
said, I think everybody likesthe hyper-personalization.
While some of the previousgeneration and even current
generation go, how do they knowso much about me?
They must be listening at thesame time.
I.
If I'm getting the samesuggestions that my 17-year-old
daughter gets, I'm gonna startto figure out what is going on
(06:49):
with this.
And I think restaurants havebeen so far behind that curve as
it relates to being able to dohyper localized suggestions
because I.
Investment in restaurants hasn'tbeen as great.
It's an offline commerce thingrather than an online commerce
thing.
So I'd love to understand whyyou think now is the time to
invest in this future, to helprestaurants, to be able to
(07:11):
excel.
'cause I think a lot of peoplehave thought about it, but I
don't necessarily think that,that the, the feed on the
pavement have been there quiteyet to get there.
So I'd love to kinda get yourthoughts and your co-founder's
thoughts on why investing time,money, and energy in this area
is the right way.
Maria Zhang (07:26):
One is, I think timing is everything, I do think
it's perfect timing, to investin, ai, automation,
hyper-personalization, liketoday.
Yeah.
So, because I think 2023.
The technology was stillimmature.
now I think it's gone through,this process of maturing both
(07:46):
the foundation, models and alsothe layers of technology on top
of it to really serve, yourguests because it's important to
always deliver a perfected guestexperience.
Yeah.
So, timing on one side istechnology, and of course on the
other side you mentioned, theconsumer expectations.
We are all used to receivingthis personalized experience and
(08:08):
we all use chat, GPT, And so,when I call the restaurant and
then I expect that instantanswer targeted to me and can
give me what I want right away,right?
I wet to be put on hold.
If it's put on hold.
If you say hold.
Jeremy Julian (08:23):
Yeah, exactly.
You'll hang up and call somebodyelse that's not gonna put you on
hold or figure out another wayto order from them.
Maria Zhang (08:27):
Yeah, because the consumer expectation has just
elevated to a point.
It also makes the adopting thistechnology necessary, right?
Because otherwise they'll bedisappointed.
And then, and then your, if yourcompetitor adopts and you don't,
and they're gonna choose to, togo there because that's the
experience that they expect and.
Jeremy Julian (08:48):
So Maria, talk to me a little bit about,'cause you
brushed over the surface of itand we've had some people on
that are really focused on voiceAI and nothing more than voice
ai and it feels like Poona isstaking a different approach.
You guys are taking amultimodality approach, which I
think is.
Very ambitious and amazing fromthe perspective of you've got
(09:10):
voice, you've got SMS, you'vegot an agent, you've got, all of
these different things.
So talk me through you've givenme an idea why you think now is
the right time to do this.
I.
From a consumer perspectivetotally makes sense.
From a restaurant perspective,they're struggling to hire
people.
Automation needs to be able tohelp them to be able to deliver
on their guest experience'causethey've got more capacity, but
don't have enough people toanswer the phones or do
(09:31):
whatever.
So why did you guys take on theambitious goal of trying to do
multimodality?
Because we've had people thatare like, and just even getting
voice right is hard.
Nonetheless, all of these otherkind of, um, transient other
areas.
Maria Zhang (09:45):
Yeah.
So, um, we are a, um, a verycommitted team, right?
We took on these challengesfully understanding the
complexity that it comes with,but we're very committed to
solve them for our customers.
and the reason why the one wordanswer is it's human likeness.
say, you give me my number.
(10:06):
Sometimes I call you, but sayI'm going through a tunnel or
something, I lost, I'm gonnatext you.
You're gonna know it's me.
And then you'll remember thecontext.
So that's exactly the experiencewe're delivering to the
restaurants as customers andtheir guests.
You don't wanna just, you canonly call, you cannot text.
Is there a friend?
Is you have a friend, you say,you can only call me.
You can never text me.
Jeremy Julian (10:27):
Yeah, exactly.
Maria Zhang (10:28):
like, you're weird.
Yeah.
and vice versa.
And also we just wanna give themflex flexibility to call and
text.
And then you made a acuteobservation.
a lot of companies offered this.
We're not the only one, but, butvery few because it's very hard
to manage the context,
Jeremy Julian (10:46):
Yep.
Maria Zhang (10:46):
So we invest a lot in both the short term memory
management and long management.
text is asynchronous, right?
Call is synchronous, I'm on thephone, right?
So how you manage memory andcontext for call and texts are
completely different.
And we just overcame, thistremendous challenge and deliver
this very seamless experience.
(11:08):
And ultimately it's a humanlikeand we really wanna meet, the
customers and the guests.
We wanna meet them wherethey're,
Jeremy Julian (11:15):
So talk me through what it is.
again, I think I've skirtedaround it, but just give our
listeners directly, we are goingto solve X and I think it's
great'cause then it's gonna addto some other questions that I
have for you about how you guysare going about it and what the
guest experience gonna look likeand the staff experience.
But ultimately, is it orders, isit just finding out if they're
(11:35):
open and the patio's open today?
Help me understand what it isthat you guys are trying to
solve for.
Maria Zhang (11:40):
The, all of the above.
Going back to the ambitious andcommitted.
so we call our agent theemployee of the century, right?
and of course we all knowemployee of the right.
Just imagine you have thisemployee.
and a phone rings.
And in he or she's gonna answerthe phone and whatever the guest
asks for, he or she'll respond.
And wouldn't be like, oh, I onlytake orders.
(12:00):
I don't, I'm not gonna tell youthe hours.
So, so that's why it has all thegeneral information.
It can make reservations, I canplace order, you can track order
status.
We.
Process, right?
We really wanted to, create thisvirtual experience, um, that you
could say, oh, this is the bestemployee I could ever have.
And of course this employeewould work 24 7 concern 500 guys
(12:23):
simultaneously.
I would never call you in a sickday.
so a lot of, the productdecisions such as why should we
support both, voice Texas,because.
Jeremy Julian (12:32):
I love that.
So how do you get around thewhole idea of we're in the
hospitality industry and a lotof people that are listening are
gonna say, yeah, a computer'snot gonna be able to do it, as
well as Susie, I.
You know who, I just trained onhow to do this and I know the
answer, but I'd love to hearyour take and why you guys think
that this is, you already talkedabout the scale, which is great.
(12:53):
They also never call out sick,which is also amazing.
And, um, I think, as an owner,as a, as an executive, you can
tell it what to say, whereastelling Susie what to say or
Jimmy what to say.
They may or may not decide tosay it on this Tuesday'cause
they had a bad day driving towork or whatever else.
But I'd love to, to help youexplain, have you explain why in
(13:13):
the hospitality sector do youthink that this is a hospitable
way of going about doingbusiness today?
Maria Zhang (13:18):
Yep.
So, I love your question.
actually, came to this industrywith humility, not with
expertise.
Yeah.
Um, so I say I'm a, I can sayI'm a computer scientist, but
I'm a amateur chef.
Jeremy Julian (13:31):
Got ya.
Maria Zhang (13:32):
I can cook for maybe four or five people.
They're decent.
They're decent, but I wouldn'tsay, I know, what hospitality
means.
Right?
And so, so the product is twopart, right?
One is the guest experience, andit just feels like a virtual
employee who's serving theguest.
Um, and with voice, with text,but on the.
(13:52):
of the product.
Um, we actually call it agentAcademy.
And this is the product for therestaurant owners.
The restaurant owners will trainthe AI agent.
Yeah.
Um, and this is also, um, theadvancement of the generative AI
technology to have the abilityto learn.
as you interact with chat GPT,you can tell, you can morph
(14:14):
check gt a personality you want.
You can give them instructions,right?
And you can say, write thislonger.
Write shorter.
You see the, this new, advancedAI model has this capability of
listening and taking feedbackand adapt to your command.
And so we.
we also tongue in cheek createdthe name like Agent Academy, and
then the restaurant owners candefine what hospitality mean for
(14:38):
your restaurant.
a fine dining restaurant andthen the Qi store restaurant and
then, um, and then, and someethnic restaurant.
And that they all have differentinterpretations.
For example, our, client pizzaon my heart, um, west Coast, you
probably heard of them.
I know you are also fromCalifornia.
on the coast of California isSurfers pizza, right?
Jeremy Julian (14:59):
Yep.
Maria Zhang (15:00):
everywhere.
They used to have TV ads, thissurfer running around.
And so defined
Jeremy Julian (15:05):
I.
Maria Zhang (15:06):
Jimmy, the surfer, and when you call them, visit
Jimmy and.
if it's, Hey dude, it might notbe
Jeremy Julian (15:13):
It is not appropriately for Morton
Steakhouse.
Probably, yeah.
Maria Zhang (15:16):
exactly right.
So, so we actually put therestaurant owners on the
driver's seat, we justfacilitated the technology, the
flexibility, the, to be adaptiveand really serve every brand.
Is.
Jeremy Julian (15:30):
Yes.
Maria Zhang (15:31):
we right now have a focus on Pizza because Wired
magazine covered us as, yourpizza guy is now ai and it just
naturally got a lot of attentionin pizza.
And then I was challenged andthey say Chicago Pizza is
American Pizza.
And then if you wanna sellChicago pizza, you have to have
the Chicago accent.
Jeremy Julian (15:49):
Yes.
Maria Zhang (15:50):
and honestly I was like, what is Chicago accent?
But you can, I wanna invite youto go out to our website on po
FB Food and Bev.
Yeah, there's a link there.
And then we actually have asynthesized, Chicago accent.
and um, and hopefully you'llfind a satisfactory.
Jeremy Julian (16:07):
it's funny that you say that because I think
it's critical.
I was talking to somebody at,um, big Chicken, which happens
to be owned by Shaquille O'Neal,and they were gonna train the AI
to have Shaquille answer thephone, and they were talking
about Colonel Sanders answeringthe phone for Kentucky Fried
Chicken.
And I think it's, um, it's a,it's remarkable, um, that the
tools have gotten so great.
(16:28):
Even two years ago, you said,it's 2025 when we're recording
this, 20, even in 2023.
The AI was not there.
It was too slow.
It didn't, it, for a real liveconversation, it wasn't there.
And in just two short years I'vebeen blown away when I play with
tools like Poona where I log inand I'm like, okay, let me order
a cheeseburger.
(16:49):
And it walking me through iteven better and giving me more
options than.
Um, than even, a staff membermight.
And so, um, one of the otherthings, Maria, that I'd love
your opinion on, and I heardthis last week at.
Um, at RLC it's really the, thisconversation of the human puts
(17:10):
their own worldview in front ofthe guest every time they come
at it, versus the owner puttingtheir worldview in front of the
guest when you're usingtechnology to communicate.
I'd love your thoughts on that.
and by that I'm saying, you knowwhat?
They've got a worldview thatsays somebody that has an accent
or somebody that's heavy, orsomebody that's, has children,
(17:30):
they're going to take their ownworldview and maybe offer them
something that they like ordon't like that they think the
kids might like.
And so versus the owner saying.
Everybody should get offered theshake regardless of whether
they're, they're here or there,they're there.
Everybody should be offered theshake.
Everybody should be offered asample cookie, whatever that
might be, because it'sconsistent and it comes with a
(17:51):
brand promise.
I'd love your thoughts on how AIand tools like Poona can create
that standardized guestexperience to get the restaurant
to do what they're looking todo.
And now a word from one of oursponsors.
Every restaurant operatorunderstands the chaos of a
restaurant kitchen during themeal rush.
Restaurant technologies, oil,total oil management solutions,
an end-to-end automated oilmanagement system that delivers
(18:14):
filters, monitors, and recyclesyour cooking oil, taking the
dirtiest jobs outta your kitchenand letting your employees focus
on more important tasks.
Control the kitchen, chaos withrestaurant technologies, and
make your kitchen safer.
No upfront cost.
To learn more, check out RTIdash i c.com or call 8 8 8 7 9 6
(18:35):
4 9 9 6.
Maria Zhang (18:36):
Yep.
Yep.
Um, I love how you described,the situation.
On one side, we wantconsistency, right?
we don't want, one staff membertalk about the product companies
way the other one, and they, itjust becomes very chaotic.
And so.
think step one is a standard,process and some consistency and
(18:57):
the brand voice and brand valueand so forth.
but looking forward, right?
I think the standard actually isbecoming a hyper personalized,
individualized experience,right?
And not a universal experience.
Um, so that's why, um, on theexperience side, it, one is it
(19:18):
needs to be two way street.
I call the conversationalmarketing
Jeremy Julian (19:21):
Yep.
Maria Zhang (19:22):
Versus everybody get 20% off.
does conversational marketingwork?
let me share with you, this isthe real story I have blew my
mind and AI extensive.
React and then I have to tellyou, the reception are better
than I, expected, right?
We have a little video of theseuser studies I call a surprise
(19:43):
and be lights, right?
They were just laughing and theywere just like, so this, the
marketing message, instead of20% off whatever it becomes,
happy Friday or TGI Friday.
your plan for the weekend?
It's like you care about, andthis is message was sent to
existing, customers that alreadyhave a relationship with the
restaurant, right?
(20:03):
you would see people responding.
say, I got no plans, right?
And then of course the AI say.
Let's throw a party, let's ordersome chicken, get some I'll hook
you up.
And then the other person soundsgood.
Let's do right.
And then the funniest one wasthis other person said, oh, um,
my, my Saturday, my son hassoccer.
(20:24):
Yeah.
And it like, awesome, great.
Weather's gonna be not whatever.
Just like you're conversing.
And then the guy said, how aboutI order 10 pizza, for the kids
for the game?
Yeah.
And then, and then the agentsaid, that's great.
You get 20% off.
They actually running apromotion for 20% off.
But I'm ordering temp pizza.
can I have 30%?
I was like, basic, my customeris haggling
Jeremy Julian (20:44):
Haggling with the ai.
Maria Zhang (20:46):
did give him the 30%, but can you imagine how
happy he would be when he gottemp pizza?
And he is like sharing, so he'sgonna tell oh look, I won.
it's gamified.
He earned that 30% right?
Jeremy Julian (21:00):
I love that.
That's a really cool, that's areally cool example.
Maria Zhang (21:04):
But this is marketing, we give promotions,
we give discounts of, weincentivize a behavior, but it's
so hyper personalized.
And then for the individual, theguest feels like This restaurant
is taking care of me, right?
This, the restaurant knows meand understands me and is taking
care of me, and I got this.
Jeremy Julian (21:24):
it's,
Maria Zhang (21:24):
running a.
Jeremy Julian (21:25):
and it's funny that you say that,'cause one of
the things that I do often whenI go out to eat is if I'm
between two menu items, I willask the service staff member,
Hey, I'm thinking about thechicken and the salmon.
And I ask them, you see it comeoutta the kitchen every day, but
the AI has the data.
How many people are ordering thesalmon?
How many people are ordering thechicken?
How many people are sending thechicken back?
How many of these people aresending the salmon back?
(21:46):
What is the throughput?
What are the star ratings?
What happens on Google reviewsand Yelp for this salmon versus
the chicken?
Right now, I do all of thatmyself before I go out to eat.
I was just dining last week inAtlanta at a restaurant, and I
asked the server and theserver's oh, no.
This is the one that everybodygets.
at the end of the day it was agood item, but I'm still now
(22:06):
left wondering, and I know thatAI can start to do that.
Whether I'm in chat or I'm invoice or I'm in SMS, it can say,
Hey, you normally get, younormally like items that look
like this, that feel like this'cause it knows who I am and it
knows what the restaurant.
Would you like this new specialbecause it's coming out geared
towards people like you.
Is that kind of where you guysare seeing the future for this,
(22:28):
with the suggestive ordering andsuch?
I.
Maria Zhang (22:29):
Absolutely.
And I think the future is nowlike it's happening already.
Jeremy Julian (22:34):
I love that.
So how do you guys go aboutdeploying?
I mean we're getting I'd love tounderstand because the large
language models, when I'mtalking about, who's gonna win
the Kentucky Derby or when's thebest time to book a flight to go
to?
To Dallas.
All.
I don't wanna say that's allkind of general information, and
I think with restaurants,because it is so
hyper-personalized, it's hard.
(22:55):
It's hard to get the data intothe tool in a way that you need
to.
It's gonna continue to learn.
But help me understand what theonboarding process looks like.
If I'm a restaurant and I wannastart to deploy solution like
this.
Maria Zhang (23:05):
Yeah.
Yeah.
It's actually super simple.
Um, so we all know how busy,restaurant owners and operators
are, so we definitely don't wantto create new work, right?
Um, so one thing is we did notcreate a new workflow.
We just get into the existingworkflow, right?
Because right now you eitherhave a staff member and also a
(23:26):
lot call.
You, you have a human beinganswering a call.
And so now basically that callgets forwarded, just a very
simple call forwarding,forwarding to the AI agent.
Yeah.
And the AI agent would answer,and then whatever the, the
human, the employee, staff, orcall center do, the AI agent,
the agent part is performstasks.
(23:46):
It would do exactly the.
and then we'll send, paymentgoing through the, to the end
user.
so there, this is one thing Ilike to share with your
listeners.
Um, let the AI agent process thepayment first before you send
the ticket to the
Jeremy Julian (24:00):
Yes.
Maria Zhang (24:01):
Because initially we served, That, that we just
say, oh, you can come there.
Were just a higher number ofnohow.
and then we do not want no-showsbecause it definitely just hurts
margin.
and this is just human nature.
They know it's ai.
They feel like they were justmessing around the way that they
don't feel morally obligated togo pick up the pizza.
But if you like, so, so, so nowwe change the workflow.
(24:24):
Validated and then the ordergoes to the.
Jeremy Julian (24:27):
It is not too dissimilar than 20 years ago
when they started onlineordering.
People would go order 20 pizzasor they'd make a phone call and
they wouldn't come pick'em up,and then they'd be frustrated
'cause they didn't get that.
Maria Zhang (24:37):
yeah.
But we did see a uptake on theno shows when we first, o of
course you want to, print anorder, so, so they don't have
wait or whatever.
But no, just payment first.
And then I think all the onlineorders you have to pay,
Jeremy Julian (24:49):
Yeah, nowadays you do it.
Maria Zhang (24:51):
there's no cash on delivery anymore.
You gotta pay.
Jeremy Julian (24:53):
huh.
Yeah.
No, but I'm old enough toremember a day when I could just
go online order, it would sendthrough a fax to their store.
They'd make my pizza for me, andthen I'd pay when I got there.
Um, so I love that you guys arecoming inline because that's one
of the things that a lot ofthese people, they want to try
these types of technologies, butthey struggle.
Talk to me a little bit, Maria,about guest identification
(25:14):
because in an offline commerceworld, on my Google, on my
Facebook, on my.
Microsoft, it knows who I ambecause I'm logged into my
browser or I'm logged into myapp or whatever else in
restaurants.
It's hard to know.
And so identifying who thatperson is, whether I'm a chat
bot, two of them, or I'm SMS orI'm calling, all of those are
different modalities.
And so how do you guys have adatabase to tie those things all
(25:35):
together?
Maria Zhang (25:36):
Yeah, we have a very simplistic approach.
it's the phone number.
We actually don't introduceanother id.
it's just, it's always the phonenumber.
If this is your phone number,this is you.
And I think generally it'ssocially accepted.
And of course, we're compliant.
People can opt out of it, andthey, if they opt out and they
opt out.
But if they, choose to Ian, andthen they're in the system.
(26:00):
name, your orders, yourpreference, and so forth.
Jeremy Julian (26:03):
Love that.
Um, real quick, just'cause I'dlove your opinion on it, and
I've heard this from somedifferent technologists in the
spaces, is how do you get the,um, AI from hallucinating, I
guess from the perspective of Icall the pizza place and now
it's offering me, burgers andobviously if I don't serve
burgers and I only soap sharepizzas, you've gotta be able to
(26:24):
put guardrails around thesebusiness systems and a B2C if
perplexity or Claude gives methe wrong thing.
It's just me that looks stupidwhen I post that online.
But in a restaurant environment,in a B2C environment, it's hard.
You gotta make sure that you putthose protections on.
I'd love to help you understand.
Maria Zhang (26:38):
Yep.
This is a very, challengingproblem.
And then this is just a littlebit of a cautionary tale for
buyers, right?
definitely, do your duediligence.
definitely, test the product,try to put pressure on it.
Um, yeah.
So, wouldn't say the solution isa hundred percent secure, but,
another couple question, to askis, um, is, one is grounding,
(27:00):
right?
So anti hall, the antidotes isgrounding, right?
Grounding basically is,connecting to facts, so you ask
is your large language modelgrounded?
And if they say, I, I don't knowwhat you mean.
They probably didn't do it.
other key, key word is a, redteam.
So we actually invest a lot inred teaming.
so red teaming, this termoriginally came from, um,
(27:22):
publication journalism.
Yeah.
And then was borrowed ortransferred to large language
models.
Um, um, open, yeah.
Has significant investment inred, you mean Basically you try
to break it.
Jeremy Julian (27:35):
Yes.
Maria Zhang (27:35):
come up with all the scenarios that it could
break right.
To see if it has broken.
Yeah.
So, so this is also just a lotof investment from a technology
perspective, that you have to gothrough.
and then the last one, um, I'veseen it happen is jailbreaking,
right?
so they.
Attackers and they try tojailbreak.
(27:56):
Yeah.
So, um, we definitely have donea lot of these, these tests
ourselves, right?
To make sure our models are notjailbroken and so on and so
forth.
And jailbroken.
It could be, and I think, justfood and beverage order.
It could be a target, right?
And then these smart hackers andthey just mess around and they
will post on Reddit and they.
your order agent started givingup pizza or you just, it can't
(28:19):
be quite dangerous.
So I do wanna caution, buyers todo these verification before you
commit, because they could have,both very material, um, loss of
revenue and.
And also reputational.
Jeremy Julian (28:32):
Yep.
Maria Zhang (28:32):
started curse words or whatever they can.
Jail started hallucinating bad.
Jeremy Julian (28:38):
thank you for digging into that.
'cause I've just heard, and Iknow that you have to put
guardrails, especially in a B2C'cause you can impact a brand's
image based on what's going onwithin the tech.
Um, last line of questioning,just'cause I've heard this from
time to time, and I'd love justhow you guys have dealt with
this, is, I don't wanna saythose off menu items, but those
items that I call the restaurantbecause they have all the
(29:01):
components, but it may not be ontheir menu.
It may be one of those thingsthat I say, Hey, you know what?
My kid really wants a hot dog,cutup, and in their mac and
cheese from your restaurant.
We've got mac and cheese, we'vegot hot dogs, but they're not a
menu item.
And so, um, some people I know,for myself, um, we'll call
restaurants when we wantsomething off menu because I
(29:21):
can't get it.
Um, I can't get it on theironline me menu.
How is AI and or some form ofAsian or is it just, hey, you
know what?
For those corner cases whereit's 1% of the items, we just
route them straight to a humanand the human deals with, those
complexities.
I'm fine with that answer.
I just would love your thoughtson that.
Maria Zhang (29:39):
we gave the decision, to the restaurant.
So, going back to Asian Academy,right?
You can just give theinstructions and say, yes, you
can take orders like this, orNo, you cannot.
And then just give, just stepby.
Step instructions.
it's called academy because theAI agent can learn, get trained,
and you can also change yourmind and redefine it all in a
very simple, just web-based ui,just with natural language.
(30:02):
You can just give these type ofinstructions.
Um, yeah, so, and then, and thenwhen we integrate with the
kitchen, display system, there'salso item.
The notes section, because.
Jeremy Julian (30:15):
This is what they're looking for.
I love that we have a lot ofpeople that just say, I mean
I've heard a lot of peopleonline ordering, text ordering,
all these different system voiceordering systems.
They're like, no.
We're just gonna keep guardrailson and we're not gonna let the
guest get what they want.
Which is not as hospitable asyou said.
And so, um, just'cause youbrought up another thought as
you were explaining that, um.
Help me understand how thelanguage model continues to
(30:37):
learn inside of the restaurant,because I think what they have
today, if they deploy today, isgonna be very different than
what they have a year from now,because it's gonna continue to
get better.
Help our listeners understandhow that all works and you
ensure that you're continuing todrive that and that the guests
are continuing to get even abetter experience now than they
would six months from now or ayear from now.
(30:57):
It's gonna continue to getbetter.
I'd love your thoughts on thatand how the system continues to
do that.
Maria Zhang (31:02):
So I think our product, we took a somewhat
unique approach, right?
I'm not gonna quote all thereinforcement this in, like all
this, computer scientific terms,but what we did, the solution
right, really is very much,similar to real life, right?
So that's why we have this agentacademy, right?
We don't assume this agent just.
(31:22):
Know we have supervisors, right?
And then so ares this employeebad being supervised.
Um.
the, basically the, the morefeedback you give to the agent
will be better unless you thinkthe agent's doing perfect
already.
Jeremy Julian (31:37):
Yeah.
Maria Zhang (31:37):
that.
We don't assume that.
And we have this, this veryeasy, UI for the restaurant
staff to, to correct the badbehaviors, tell him what to do.
And then, the other side, it'salso tremend.
Jeremy Julian (31:48):
Yes.
Maria Zhang (31:49):
So in general, even using chat, chat TP very
recently, right?
They've been working at this foryears.
Very recently improved a majorimprovement in memory
improvement.
If it's hard for, jet TBD isvery hard for us as a small
company, but we really invest inthat because think about how can
you get better.
You have to remember.
If you don't remember, like Ican tell you chat GB 3.5, didn't
(32:12):
remember nothing, remember much.
the new one is much better.
But you have to make the agent,remember each and every guest,
right?
That's how you perfect thatexperience.
Jeremy Julian (32:22):
Yeah.
Maria Zhang (32:23):
it's, we're not perfect, but we definitely
recognize how important, and howchallenging this is.
And is our approach.
Yeah.
Jeremy Julian (32:30):
I love that.
Maria, I could talk all day,but, why don't you give our
listeners, I, it fascinates me.
I'm blown away.
I continue, I watch YouTubevideos every day almost about
how, these large language modelscan make my life and our
customer's lives better.
So I love that you guys areinvesting in really.
Those end users that are havingto take the phone calls and
having to figure out all ofthis, they've got so many things
(32:52):
on their plate.
And so I love that you guys areinvesting in a solution to
really help make their livesbetter.
And so if people wanna learnmore, if they wanna, you already
threw it out, Poona ai.ai, slashfood and beverage or whatever,
but is that really where theylearn more?
Do they do, should they connectwith you?
Is there anything specific youwould direct them to, to have
them, have them get educatedabout these things and stay up
(33:12):
to date?
Maria Zhang (33:13):
absolutely.
Come visit our website and thenalso just con connect me on
LinkedIn.
and I'm also going to theNational Restaurant Show.
just drop me a note.
I am a experienced, a computerscience computer engineer, but A
passionate and amateur chef, anddefinitely a foodie.
I've really love.
This the intersection of myprofessional life and personal
(33:36):
passion.
So I love talk to restaurateurs.
I find all of you veryfascinating.
Jeremy Julian (33:42):
I love that.
Maria Zhang (33:43):
just love to just have a chat, just, linking
website or see at the, um, next,next restaurant show.
Jeremy Julian (33:51):
So what is your go-to item for, a party.
You have a dinner party at yourhouse tonight.
What are you making?
What's your go-to foodie itemthat you're making at the house?
Maria Zhang (33:59):
Um, it, it depends if I wanna be, like a hands-on,
everybody involved, I love tohost a dumpling party.
party.
Jeremy Julian (34:07):
Okay.
Maria Zhang (34:07):
make the dough and roll the skin and make the
feelings and different flavorsand have everybody make the
dumplings, It's very crafty.
so that's, that's, um.
Um, let's all roll up oursleeves and work together.
Jeremy Julian (34:21):
That sounds amazing.
I haven't gotta figure out howto do a dumpling party.
Maria Zhang (34:25):
But, but if it's just quick and easy, of course
love to order some deliciouspizza and throw some salad.
Everybody loves that.
Yeah.
So, like I said, I'm a mature so
Jeremy Julian (34:35):
I love that.
Maria, I said it earlier.
I ca I am blown away with howthe speed of technology, and
quite frankly, even the lastfive years in my life, the
technology and the advancementhas accelerated so greatly and
it's ultimately when done right,it's gonna make restaurants and
consumers' life so much easier.
So thank you guys for doingthat.
(34:55):
To our listeners guys, you guyshave got lots of choices, so
thank you guys for hanging out.
If you haven't already done so,please subscribe to the show on
your favorite listener and makeit a great day.
Maria Zhang (35:04):
Thank you so.
In today's episode, we are joined by Maria,
who is uh, one of the foundersof palona.
Pal's really solving the world'sproblems as it relates to voice
automation, and not justordering, but all of the
different items that you can usevoice for.
To deliver a better guestexperience.
(00:22):
She's got a fantastic pedigreewhere she's worked at a lot of
large brands and is using all ofthat skill and the relationships
that she has and the team thatshe has built to really change
the world of restaurantautomation and how it engages
with voice Sta until the end,where we talk a lot about how
restaurants can implement thissolution today to ultimately
(00:46):
drive bottom line profit totheir brand.
If you don't know me.
My name is Jeremy Julian.
I am the Chief Revenue Officerfor CB S North Star.
We wrote the North Star point ofsale solution for multi-units.
Check us out@cbsnorthstar.comand now onto the episode.
I.
Welcome back to the RestaurantTechnology Guys podcast.
I thank everyone out there forjoining us.
(01:07):
Like I say, each and every time,I know you guys got lots of
choices.
So thanks for hanging out.
Today is, um, we're gonna talk alot about kind of AI and really,
quite frankly, with our gueststalk a lot about.
How the consumer continues tochange and how some of the
non-restaurant tech is really,um, making a huge impact in our
(01:28):
space.
And so I'm gonna introduce youto our guests.
Maria, why don't you introduceyourself.
Let's talk a little bit aboutwhere your background is and
where you came from, and then wecan talk about what you've built
and what your latest project is.
Maria Zhang (01:40):
Yeah.
Hi Jeremy.
Thank you so much for having meon the show.
I.
From you and share a little bitfrom my story.
Yeah.
So I stu I studied computerscience and have been in this
industry, for over two decades.
Um, most recently I was VP ofengineering at Google.
I supported a large global,product development team of,
(02:00):
several hundred people.
and before that I was at meta.
And, I also have done a startup,back in, um, thousand 10.
I left Microsoft and when iPhonecame out, I got really excited
and the product I built wascalled Alike.
It's a local discovery andrecommendation tool.
So during that time, already getto work with, local restaurants
(02:22):
and such, and how to make smartrecommendations.
And that product was, acquiredby Yahoo in 2013.
So fast forward, this companyPalona.
I started with a fewco-founders, last year in 2024.
and we really excited about thisgenerative AI technology.
And then we want to bring themost advanced, multimodal,
(02:45):
multimodality generative AIsolution to serve, restaurants
and make the guest experiencebetter and make the operations,
more seamless and more costeffective.
Jeremy Julian (02:55):
Wow, you got your, you got your pitch deck
down.
I like it.
Um, as far as where it came fromand the fact that you've got
such experience, Microsoft Meta,um, Google is awesome.
I.
We were only, about 55 secondsin before AI came up.
And so Maria, I would love foryou to define what is ai,
because I think there's a lot ofpeople that it's a misnomer and
(03:16):
being in the tech space, you'vebeen in the tech space for a
long time.
There's a lot of these thingsthat have been happening for
some time now.
But I think because of.
Open AI and chat.
GPT has really put it inconsumer's hands in a way that,
I think, even five years ago orthree years ago, nobody really
knew while some of this stuffwas happening in the background
(03:36):
with, with Google and some ofthese other large companies, now
it's become almost mainstreamwhere you talk about it.
And so define what it is andwhat is this large language
model and why do you think ithas such an opportunity to solve
so many problems within therestaurant space?
Maria Zhang (03:50):
Yeah, that's a great question.
Yeah.
So, you are absolutely right.
the term right now is bothubiquitous and honestly a little
overused.
Um, but AI is not a new concept,right?
this concept started decadesago.
and I think as a.
Say 20 years ago, we avoid, usethis term artificial
(04:11):
intelligence because they soundso dystopic at all the movies we
all watch the Sky night andsuch.
And so, so we, use alternativeterms like machine learning,
predictive models.
lemme give you example.
I worked at Zillow in know veryearly days of Zi.
There's.
Long, long time ago, 2006, Ithink.
but if you think about Zillow,that is definitively artificial
(04:33):
intelligence because you could,spend what,$1,200 and hire
assessor to give you a estimatorassessment value.
Or you can go to Zillow and havethe algorithm right.
Gave you estimate, which is.
All, but then Zillow positionedthemselves as useful product,
right?
Um, so fast forward, I think,was 2022 a chat.
(04:58):
GPT came out in 2324, as you cansee, spearheaded by AI and
Microsoft copilot, Google Jamand Perplexity.
the name just layers on AppleIntelligence, right?
Amazon has a.
Is it's inflection point.
So a couple things happened.
one is, they're called largelanguage models.
and it's a descrip.
(05:18):
They're very large.
They have, many billions of,perimeters, parameters.
And what made this.
possible is, high performancecomputing.
And that's why Nvidia stock isso high because of GPU.
And then there was anotheralgorithm called Transformer
that really empowered, massiveparallel computing.
(05:39):
So this is another, Of computerscience called high performance
computing.
So it's h PC that empowered verylarge models that enabled all
these capabilities.
Yeah, so, so they're generative,right?
Meaning they can, not onlyreturn a paragraph, but even a
article even, you can write anovel if you want, if you wanna
read that novel, a large amountof text, This is I mentioned
(06:03):
earlier is multimodality.
Jeremy Julian (06:06):
and I think even going back to some of your early
days, it's amazing becauseagain, it's been around, we
think about the friendsuggestions that are on your
Facebook feed or on your Twitterfeed, or just even what shows up
in your feed.
When I sit next to my wife onthe couch and we're both
scrolling, Instagram, herInstagram feed looks very
different than mine when I loginto Gmail.
Um, her Gmail looks verydifferent than mine, and I
(06:28):
think.
Because the world of compute hasbecome so ubiquitous, as you
said, I think everybody likesthe hyper-personalization.
While some of the previousgeneration and even current
generation go, how do they knowso much about me?
They must be listening at thesame time.
I.
If I'm getting the samesuggestions that my 17-year-old
daughter gets, I'm gonna startto figure out what is going on
(06:49):
with this.
And I think restaurants havebeen so far behind that curve as
it relates to being able to dohyper localized suggestions
because I.
Investment in restaurants hasn'tbeen as great.
It's an offline commerce thingrather than an online commerce
thing.
So I'd love to understand whyyou think now is the time to
invest in this future, to helprestaurants, to be able to
(07:11):
excel.
'cause I think a lot of peoplehave thought about it, but I
don't necessarily think that,that the, the feed on the
pavement have been there quiteyet to get there.
So I'd love to kinda get yourthoughts and your co-founder's
thoughts on why investing time,money, and energy in this area
is the right way.
Maria Zhang (07:26):
One is, I think timing is everything, I do think
it's perfect timing, to investin, ai, automation,
hyper-personalization, liketoday.
Yeah.
So, because I think 2023.
The technology was stillimmature.
now I think it's gone through,this process of maturing both
(07:46):
the foundation, models and alsothe layers of technology on top
of it to really serve, yourguests because it's important to
always deliver a perfected guestexperience.
Yeah.
So, timing on one side istechnology, and of course on the
other side you mentioned, theconsumer expectations.
We are all used to receivingthis personalized experience and
(08:08):
we all use chat, GPT, And so,when I call the restaurant and
then I expect that instantanswer targeted to me and can
give me what I want right away,right?
I wet to be put on hold.
If it's put on hold.
If you say hold.
Jeremy Julian (08:23):
Yeah, exactly.
You'll hang up and call somebodyelse that's not gonna put you on
hold or figure out another wayto order from them.
Maria Zhang (08:27):
Yeah, because the consumer expectation has just
elevated to a point.
It also makes the adopting thistechnology necessary, right?
Because otherwise they'll bedisappointed.
And then, and then your, if yourcompetitor adopts and you don't,
and they're gonna choose to, togo there because that's the
experience that they expect and.
Jeremy Julian (08:48):
So Maria, talk to me a little bit about,'cause you
brushed over the surface of itand we've had some people on
that are really focused on voiceAI and nothing more than voice
ai and it feels like Poona isstaking a different approach.
You guys are taking amultimodality approach, which I
think is.
Very ambitious and amazing fromthe perspective of you've got
(09:10):
voice, you've got SMS, you'vegot an agent, you've got, all of
these different things.
So talk me through you've givenme an idea why you think now is
the right time to do this.
I.
From a consumer perspectivetotally makes sense.
From a restaurant perspective,they're struggling to hire
people.
Automation needs to be able tohelp them to be able to deliver
on their guest experience'causethey've got more capacity, but
don't have enough people toanswer the phones or do
(09:31):
whatever.
So why did you guys take on theambitious goal of trying to do
multimodality?
Because we've had people thatare like, and just even getting
voice right is hard.
Nonetheless, all of these otherkind of, um, transient other
areas.
Maria Zhang (09:45):
Yeah.
So, um, we are a, um, a verycommitted team, right?
We took on these challengesfully understanding the
complexity that it comes with,but we're very committed to
solve them for our customers.
and the reason why the one wordanswer is it's human likeness.
say, you give me my number.
(10:06):
Sometimes I call you, but sayI'm going through a tunnel or
something, I lost, I'm gonnatext you.
You're gonna know it's me.
And then you'll remember thecontext.
So that's exactly the experiencewe're delivering to the
restaurants as customers andtheir guests.
You don't wanna just, you canonly call, you cannot text.
Is there a friend?
Is you have a friend, you say,you can only call me.
You can never text me.
Jeremy Julian (10:27):
Yeah, exactly.
Maria Zhang (10:28):
like, you're weird.
Yeah.
and vice versa.
And also we just wanna give themflex flexibility to call and
text.
And then you made a acuteobservation.
a lot of companies offered this.
We're not the only one, but, butvery few because it's very hard
to manage the context,
Jeremy Julian (10:46):
Yep.
Maria Zhang (10:46):
So we invest a lot in both the short term memory
management and long management.
text is asynchronous, right?
Call is synchronous, I'm on thephone, right?
So how you manage memory andcontext for call and texts are
completely different.
And we just overcame, thistremendous challenge and deliver
this very seamless experience.
(11:08):
And ultimately it's a humanlikeand we really wanna meet, the
customers and the guests.
We wanna meet them wherethey're,
Jeremy Julian (11:15):
So talk me through what it is.
again, I think I've skirtedaround it, but just give our
listeners directly, we are goingto solve X and I think it's
great'cause then it's gonna addto some other questions that I
have for you about how you guysare going about it and what the
guest experience gonna look likeand the staff experience.
But ultimately, is it orders, isit just finding out if they're
(11:35):
open and the patio's open today?
Help me understand what it isthat you guys are trying to
solve for.
Maria Zhang (11:40):
The, all of the above.
Going back to the ambitious andcommitted.
so we call our agent theemployee of the century, right?
and of course we all knowemployee of the right.
Just imagine you have thisemployee.
and a phone rings.
And in he or she's gonna answerthe phone and whatever the guest
asks for, he or she'll respond.
And wouldn't be like, oh, I onlytake orders.
(12:00):
I don't, I'm not gonna tell youthe hours.
So, so that's why it has all thegeneral information.
It can make reservations, I canplace order, you can track order
status.
We.
Process, right?
We really wanted to, create thisvirtual experience, um, that you
could say, oh, this is the bestemployee I could ever have.
And of course this employeewould work 24 7 concern 500 guys
(12:23):
simultaneously.
I would never call you in a sickday.
so a lot of, the productdecisions such as why should we
support both, voice Texas,because.
Jeremy Julian (12:32):
I love that.
So how do you get around thewhole idea of we're in the
hospitality industry and a lotof people that are listening are
gonna say, yeah, a computer'snot gonna be able to do it, as
well as Susie, I.
You know who, I just trained onhow to do this and I know the
answer, but I'd love to hearyour take and why you guys think
that this is, you already talkedabout the scale, which is great.
(12:53):
They also never call out sick,which is also amazing.
And, um, I think, as an owner,as a, as an executive, you can
tell it what to say, whereastelling Susie what to say or
Jimmy what to say.
They may or may not decide tosay it on this Tuesday'cause
they had a bad day driving towork or whatever else.
But I'd love to, to help youexplain, have you explain why in
(13:13):
the hospitality sector do youthink that this is a hospitable
way of going about doingbusiness today?
Maria Zhang (13:18):
Yep.
So, I love your question.
actually, came to this industrywith humility, not with
expertise.
Yeah.
Um, so I say I'm a, I can sayI'm a computer scientist, but
I'm a amateur chef.
Jeremy Julian (13:31):
Got ya.
Maria Zhang (13:32):
I can cook for maybe four or five people.
They're decent.
They're decent, but I wouldn'tsay, I know, what hospitality
means.
Right?
And so, so the product is twopart, right?
One is the guest experience, andit just feels like a virtual
employee who's serving theguest.
Um, and with voice, with text,but on the.
(13:52):
of the product.
Um, we actually call it agentAcademy.
And this is the product for therestaurant owners.
The restaurant owners will trainthe AI agent.
Yeah.
Um, and this is also, um, theadvancement of the generative AI
technology to have the abilityto learn.
as you interact with chat GPT,you can tell, you can morph
(14:14):
check gt a personality you want.
You can give them instructions,right?
And you can say, write thislonger.
Write shorter.
You see the, this new, advancedAI model has this capability of
listening and taking feedbackand adapt to your command.
And so we.
we also tongue in cheek createdthe name like Agent Academy, and
then the restaurant owners candefine what hospitality mean for
(14:38):
your restaurant.
a fine dining restaurant andthen the Qi store restaurant and
then, um, and then, and someethnic restaurant.
And that they all have differentinterpretations.
For example, our, client pizzaon my heart, um, west Coast, you
probably heard of them.
I know you are also fromCalifornia.
on the coast of California isSurfers pizza, right?
Jeremy Julian (14:59):
Yep.
Maria Zhang (15:00):
everywhere.
They used to have TV ads, thissurfer running around.
And so defined
Jeremy Julian (15:05):
I.
Maria Zhang (15:06):
Jimmy, the surfer, and when you call them, visit
Jimmy and.
if it's, Hey dude, it might notbe
Jeremy Julian (15:13):
It is not appropriately for Morton
Steakhouse.
Probably, yeah.
Maria Zhang (15:16):
exactly right.
So, so we actually put therestaurant owners on the
driver's seat, we justfacilitated the technology, the
flexibility, the, to be adaptiveand really serve every brand.
Is.
Jeremy Julian (15:30):
Yes.
Maria Zhang (15:31):
we right now have a focus on Pizza because Wired
magazine covered us as, yourpizza guy is now ai and it just
naturally got a lot of attentionin pizza.
And then I was challenged andthey say Chicago Pizza is
American Pizza.
And then if you wanna sellChicago pizza, you have to have
the Chicago accent.
Jeremy Julian (15:49):
Yes.
Maria Zhang (15:50):
and honestly I was like, what is Chicago accent?
But you can, I wanna invite youto go out to our website on po
FB Food and Bev.
Yeah, there's a link there.
And then we actually have asynthesized, Chicago accent.
and um, and hopefully you'llfind a satisfactory.
Jeremy Julian (16:07):
it's funny that you say that because I think
it's critical.
I was talking to somebody at,um, big Chicken, which happens
to be owned by Shaquille O'Neal,and they were gonna train the AI
to have Shaquille answer thephone, and they were talking
about Colonel Sanders answeringthe phone for Kentucky Fried
Chicken.
And I think it's, um, it's a,it's remarkable, um, that the
tools have gotten so great.
(16:28):
Even two years ago, you said,it's 2025 when we're recording
this, 20, even in 2023.
The AI was not there.
It was too slow.
It didn't, it, for a real liveconversation, it wasn't there.
And in just two short years I'vebeen blown away when I play with
tools like Poona where I log inand I'm like, okay, let me order
a cheeseburger.
(16:49):
And it walking me through iteven better and giving me more
options than.
Um, than even, a staff membermight.
And so, um, one of the otherthings, Maria, that I'd love
your opinion on, and I heardthis last week at.
Um, at RLC it's really the, thisconversation of the human puts
(17:10):
their own worldview in front ofthe guest every time they come
at it, versus the owner puttingtheir worldview in front of the
guest when you're usingtechnology to communicate.
I'd love your thoughts on that.
and by that I'm saying, you knowwhat?
They've got a worldview thatsays somebody that has an accent
or somebody that's heavy, orsomebody that's, has children,
(17:30):
they're going to take their ownworldview and maybe offer them
something that they like ordon't like that they think the
kids might like.
And so versus the owner saying.
Everybody should get offered theshake regardless of whether
they're, they're here or there,they're there.
Everybody should be offered theshake.
Everybody should be offered asample cookie, whatever that
might be, because it'sconsistent and it comes with a
(17:51):
brand promise.
I'd love your thoughts on how AIand tools like Poona can create
that standardized guestexperience to get the restaurant
to do what they're looking todo.
And now a word from one of oursponsors.
Every restaurant operatorunderstands the chaos of a
restaurant kitchen during themeal rush.
Restaurant technologies, oil,total oil management solutions,
an end-to-end automated oilmanagement system that delivers
(18:14):
filters, monitors, and recyclesyour cooking oil, taking the
dirtiest jobs outta your kitchenand letting your employees focus
on more important tasks.
Control the kitchen, chaos withrestaurant technologies, and
make your kitchen safer.
No upfront cost.
To learn more, check out RTIdash i c.com or call 8 8 8 7 9 6
(18:35):
4 9 9 6.
Maria Zhang (18:36):
Yep.
Yep.
Um, I love how you described,the situation.
On one side, we wantconsistency, right?
we don't want, one staff membertalk about the product companies
way the other one, and they, itjust becomes very chaotic.
And so.
think step one is a standard,process and some consistency and
(18:57):
the brand voice and brand valueand so forth.
but looking forward, right?
I think the standard actually isbecoming a hyper personalized,
individualized experience,right?
And not a universal experience.
Um, so that's why, um, on theexperience side, it, one is it
(19:18):
needs to be two way street.
I call the conversationalmarketing
Jeremy Julian (19:21):
Yep.
Maria Zhang (19:22):
Versus everybody get 20% off.
does conversational marketingwork?
let me share with you, this isthe real story I have blew my
mind and AI extensive.
React and then I have to tellyou, the reception are better
than I, expected, right?
We have a little video of theseuser studies I call a surprise
(19:43):
and be lights, right?
They were just laughing and theywere just like, so this, the
marketing message, instead of20% off whatever it becomes,
happy Friday or TGI Friday.
your plan for the weekend?
It's like you care about, andthis is message was sent to
existing, customers that alreadyhave a relationship with the
restaurant, right?
(20:03):
you would see people responding.
say, I got no plans, right?
And then of course the AI say.
Let's throw a party, let's ordersome chicken, get some I'll hook
you up.
And then the other person soundsgood.
Let's do right.
And then the funniest one wasthis other person said, oh, um,
my, my Saturday, my son hassoccer.
(20:24):
Yeah.
And it like, awesome, great.
Weather's gonna be not whatever.
Just like you're conversing.
And then the guy said, how aboutI order 10 pizza, for the kids
for the game?
Yeah.
And then, and then the agentsaid, that's great.
You get 20% off.
They actually running apromotion for 20% off.
But I'm ordering temp pizza.
can I have 30%?
I was like, basic, my customeris haggling
Jeremy Julian (20:44):
Haggling with the ai.
Maria Zhang (20:46):
did give him the 30%, but can you imagine how
happy he would be when he gottemp pizza?
And he is like sharing, so he'sgonna tell oh look, I won.
it's gamified.
He earned that 30% right?
Jeremy Julian (21:00):
I love that.
That's a really cool, that's areally cool example.
Maria Zhang (21:04):
But this is marketing, we give promotions,
we give discounts of, weincentivize a behavior, but it's
so hyper personalized.
And then for the individual, theguest feels like This restaurant
is taking care of me, right?
This, the restaurant knows meand understands me and is taking
care of me, and I got this.
Jeremy Julian (21:24):
it's,
Maria Zhang (21:24):
running a.
Jeremy Julian (21:25):
and it's funny that you say that,'cause one of
the things that I do often whenI go out to eat is if I'm
between two menu items, I willask the service staff member,
Hey, I'm thinking about thechicken and the salmon.
And I ask them, you see it comeoutta the kitchen every day, but
the AI has the data.
How many people are ordering thesalmon?
How many people are ordering thechicken?
How many people are sending thechicken back?
How many of these people aresending the salmon back?
(21:46):
What is the throughput?
What are the star ratings?
What happens on Google reviewsand Yelp for this salmon versus
the chicken?
Right now, I do all of thatmyself before I go out to eat.
I was just dining last week inAtlanta at a restaurant, and I
asked the server and theserver's oh, no.
This is the one that everybodygets.
at the end of the day it was agood item, but I'm still now
(22:06):
left wondering, and I know thatAI can start to do that.
Whether I'm in chat or I'm invoice or I'm in SMS, it can say,
Hey, you normally get, younormally like items that look
like this, that feel like this'cause it knows who I am and it
knows what the restaurant.
Would you like this new specialbecause it's coming out geared
towards people like you.
Is that kind of where you guysare seeing the future for this,
(22:28):
with the suggestive ordering andsuch?
I.
Maria Zhang (22:29):
Absolutely.
And I think the future is nowlike it's happening already.
Jeremy Julian (22:34):
I love that.
So how do you guys go aboutdeploying?
I mean we're getting I'd love tounderstand because the large
language models, when I'mtalking about, who's gonna win
the Kentucky Derby or when's thebest time to book a flight to go
to?
To Dallas.
All.
I don't wanna say that's allkind of general information, and
I think with restaurants,because it is so
hyper-personalized, it's hard.
(22:55):
It's hard to get the data intothe tool in a way that you need
to.
It's gonna continue to learn.
But help me understand what theonboarding process looks like.
If I'm a restaurant and I wannastart to deploy solution like
this.
Maria Zhang (23:05):
Yeah.
Yeah.
It's actually super simple.
Um, so we all know how busy,restaurant owners and operators
are, so we definitely don't wantto create new work, right?
Um, so one thing is we did notcreate a new workflow.
We just get into the existingworkflow, right?
Because right now you eitherhave a staff member and also a
(23:26):
lot call.
You, you have a human beinganswering a call.
And so now basically that callgets forwarded, just a very
simple call forwarding,forwarding to the AI agent.
Yeah.
And the AI agent would answer,and then whatever the, the
human, the employee, staff, orcall center do, the AI agent,
the agent part is performstasks.
(23:46):
It would do exactly the.
and then we'll send, paymentgoing through the, to the end
user.
so there, this is one thing Ilike to share with your
listeners.
Um, let the AI agent process thepayment first before you send
the ticket to the
Jeremy Julian (24:00):
Yes.
Maria Zhang (24:01):
Because initially we served, That, that we just
say, oh, you can come there.
Were just a higher number ofnohow.
and then we do not want no-showsbecause it definitely just hurts
margin.
and this is just human nature.
They know it's ai.
They feel like they were justmessing around the way that they
don't feel morally obligated togo pick up the pizza.
But if you like, so, so, so nowwe change the workflow.
(24:24):
Validated and then the ordergoes to the.
Jeremy Julian (24:27):
It is not too dissimilar than 20 years ago
when they started onlineordering.
People would go order 20 pizzasor they'd make a phone call and
they wouldn't come pick'em up,and then they'd be frustrated
'cause they didn't get that.
Maria Zhang (24:37):
yeah.
But we did see a uptake on theno shows when we first, o of
course you want to, print anorder, so, so they don't have
wait or whatever.
But no, just payment first.
And then I think all the onlineorders you have to pay,
Jeremy Julian (24:49):
Yeah, nowadays you do it.
Maria Zhang (24:51):
there's no cash on delivery anymore.
You gotta pay.
Jeremy Julian (24:53):
huh.
Yeah.
No, but I'm old enough toremember a day when I could just
go online order, it would sendthrough a fax to their store.
They'd make my pizza for me, andthen I'd pay when I got there.
Um, so I love that you guys arecoming inline because that's one
of the things that a lot ofthese people, they want to try
these types of technologies, butthey struggle.
Talk to me a little bit, Maria,about guest identification
(25:14):
because in an offline commerceworld, on my Google, on my
Facebook, on my.
Microsoft, it knows who I ambecause I'm logged into my
browser or I'm logged into myapp or whatever else in
restaurants.
It's hard to know.
And so identifying who thatperson is, whether I'm a chat
bot, two of them, or I'm SMS orI'm calling, all of those are
different modalities.
And so how do you guys have adatabase to tie those things all
(25:35):
together?
Maria Zhang (25:36):
Yeah, we have a very simplistic approach.
it's the phone number.
We actually don't introduceanother id.
it's just, it's always the phonenumber.
If this is your phone number,this is you.
And I think generally it'ssocially accepted.
And of course, we're compliant.
People can opt out of it, andthey, if they opt out and they
opt out.
But if they, choose to Ian, andthen they're in the system.
(26:00):
name, your orders, yourpreference, and so forth.
Jeremy Julian (26:03):
Love that.
Um, real quick, just'cause I'dlove your opinion on it, and
I've heard this from somedifferent technologists in the
spaces, is how do you get the,um, AI from hallucinating, I
guess from the perspective of Icall the pizza place and now
it's offering me, burgers andobviously if I don't serve
burgers and I only soap sharepizzas, you've gotta be able to
(26:24):
put guardrails around thesebusiness systems and a B2C if
perplexity or Claude gives methe wrong thing.
It's just me that looks stupidwhen I post that online.
But in a restaurant environment,in a B2C environment, it's hard.
You gotta make sure that you putthose protections on.
I'd love to help you understand.
Maria Zhang (26:38):
Yep.
This is a very, challengingproblem.
And then this is just a littlebit of a cautionary tale for
buyers, right?
definitely, do your duediligence.
definitely, test the product,try to put pressure on it.
Um, yeah.
So, wouldn't say the solution isa hundred percent secure, but,
another couple question, to askis, um, is, one is grounding,
(27:00):
right?
So anti hall, the antidotes isgrounding, right?
Grounding basically is,connecting to facts, so you ask
is your large language modelgrounded?
And if they say, I, I don't knowwhat you mean.
They probably didn't do it.
other key, key word is a, redteam.
So we actually invest a lot inred teaming.
so red teaming, this termoriginally came from, um,
(27:22):
publication journalism.
Yeah.
And then was borrowed ortransferred to large language
models.
Um, um, open, yeah.
Has significant investment inred, you mean Basically you try
to break it.
Jeremy Julian (27:35):
Yes.
Maria Zhang (27:35):
come up with all the scenarios that it could
break right.
To see if it has broken.
Yeah.
So, so this is also just a lotof investment from a technology
perspective, that you have to gothrough.
and then the last one, um, I'veseen it happen is jailbreaking,
right?
so they.
Attackers and they try tojailbreak.
(27:56):
Yeah.
So, um, we definitely have donea lot of these, these tests
ourselves, right?
To make sure our models are notjailbroken and so on and so
forth.
And jailbroken.
It could be, and I think, justfood and beverage order.
It could be a target, right?
And then these smart hackers andthey just mess around and they
will post on Reddit and they.
your order agent started givingup pizza or you just, it can't
(28:19):
be quite dangerous.
So I do wanna caution, buyers todo these verification before you
commit, because they could have,both very material, um, loss of
revenue and.
And also reputational.
Jeremy Julian (28:32):
Yep.
Maria Zhang (28:32):
started curse words or whatever they can.
Jail started hallucinating bad.
Jeremy Julian (28:38):
thank you for digging into that.
'cause I've just heard, and Iknow that you have to put
guardrails, especially in a B2C'cause you can impact a brand's
image based on what's going onwithin the tech.
Um, last line of questioning,just'cause I've heard this from
time to time, and I'd love justhow you guys have dealt with
this, is, I don't wanna saythose off menu items, but those
items that I call the restaurantbecause they have all the
(29:01):
components, but it may not be ontheir menu.
It may be one of those thingsthat I say, Hey, you know what?
My kid really wants a hot dog,cutup, and in their mac and
cheese from your restaurant.
We've got mac and cheese, we'vegot hot dogs, but they're not a
menu item.
And so, um, some people I know,for myself, um, we'll call
restaurants when we wantsomething off menu because I
(29:21):
can't get it.
Um, I can't get it on theironline me menu.
How is AI and or some form ofAsian or is it just, hey, you
know what?
For those corner cases whereit's 1% of the items, we just
route them straight to a humanand the human deals with, those
complexities.
I'm fine with that answer.
I just would love your thoughtson that.
Maria Zhang (29:39):
we gave the decision, to the restaurant.
So, going back to Asian Academy,right?
You can just give theinstructions and say, yes, you
can take orders like this, orNo, you cannot.
And then just give, just stepby.
Step instructions.
it's called academy because theAI agent can learn, get trained,
and you can also change yourmind and redefine it all in a
very simple, just web-based ui,just with natural language.
(30:02):
You can just give these type ofinstructions.
Um, yeah, so, and then, and thenwhen we integrate with the
kitchen, display system, there'salso item.
The notes section, because.
Jeremy Julian (30:15):
This is what they're looking for.
I love that we have a lot ofpeople that just say, I mean
I've heard a lot of peopleonline ordering, text ordering,
all these different system voiceordering systems.
They're like, no.
We're just gonna keep guardrailson and we're not gonna let the
guest get what they want.
Which is not as hospitable asyou said.
And so, um, just'cause youbrought up another thought as
you were explaining that, um.
Help me understand how thelanguage model continues to
(30:37):
learn inside of the restaurant,because I think what they have
today, if they deploy today, isgonna be very different than
what they have a year from now,because it's gonna continue to
get better.
Help our listeners understandhow that all works and you
ensure that you're continuing todrive that and that the guests
are continuing to get even abetter experience now than they
would six months from now or ayear from now.
(30:57):
It's gonna continue to getbetter.
I'd love your thoughts on thatand how the system continues to
do that.
Maria Zhang (31:02):
So I think our product, we took a somewhat
unique approach, right?
I'm not gonna quote all thereinforcement this in, like all
this, computer scientific terms,but what we did, the solution
right, really is very much,similar to real life, right?
So that's why we have this agentacademy, right?
We don't assume this agent just.
(31:22):
Know we have supervisors, right?
And then so ares this employeebad being supervised.
Um.
the, basically the, the morefeedback you give to the agent
will be better unless you thinkthe agent's doing perfect
already.
Jeremy Julian (31:37):
Yeah.
Maria Zhang (31:37):
that.
We don't assume that.
And we have this, this veryeasy, UI for the restaurant
staff to, to correct the badbehaviors, tell him what to do.
And then, the other side, it'salso tremend.
Jeremy Julian (31:48):
Yes.
Maria Zhang (31:49):
So in general, even using chat, chat TP very
recently, right?
They've been working at this foryears.
Very recently improved a majorimprovement in memory
improvement.
If it's hard for, jet TBD isvery hard for us as a small
company, but we really invest inthat because think about how can
you get better.
You have to remember.
If you don't remember, like Ican tell you chat GB 3.5, didn't
(32:12):
remember nothing, remember much.
the new one is much better.
But you have to make the agent,remember each and every guest,
right?
That's how you perfect thatexperience.
Jeremy Julian (32:22):
Yeah.
Maria Zhang (32:23):
it's, we're not perfect, but we definitely
recognize how important, and howchallenging this is.
And is our approach.
Yeah.
Jeremy Julian (32:30):
I love that.
Maria, I could talk all day,but, why don't you give our
listeners, I, it fascinates me.
I'm blown away.
I continue, I watch YouTubevideos every day almost about
how, these large language modelscan make my life and our
customer's lives better.
So I love that you guys areinvesting in really.
Those end users that are havingto take the phone calls and
having to figure out all ofthis, they've got so many things
(32:52):
on their plate.
And so I love that you guys areinvesting in a solution to
really help make their livesbetter.
And so if people wanna learnmore, if they wanna, you already
threw it out, Poona ai.ai, slashfood and beverage or whatever,
but is that really where theylearn more?
Do they do, should they connectwith you?
Is there anything specific youwould direct them to, to have
them, have them get educatedabout these things and stay up
(33:12):
to date?
Maria Zhang (33:13):
absolutely.
Come visit our website and thenalso just con connect me on
LinkedIn.
and I'm also going to theNational Restaurant Show.
just drop me a note.
I am a experienced, a computerscience computer engineer, but A
passionate and amateur chef, anddefinitely a foodie.
I've really love.
This the intersection of myprofessional life and personal
(33:36):
passion.
So I love talk to restaurateurs.
I find all of you veryfascinating.
Jeremy Julian (33:42):
I love that.
Maria Zhang (33:43):
just love to just have a chat, just, linking
website or see at the, um, next,next restaurant show.
Jeremy Julian (33:51):
So what is your go-to item for, a party.
You have a dinner party at yourhouse tonight.
What are you making?
What's your go-to foodie itemthat you're making at the house?
Maria Zhang (33:59):
Um, it, it depends if I wanna be, like a hands-on,
everybody involved, I love tohost a dumpling party.
party.
Jeremy Julian (34:07):
Okay.
Maria Zhang (34:07):
make the dough and roll the skin and make the
feelings and different flavorsand have everybody make the
dumplings, It's very crafty.
so that's, that's, um.
Um, let's all roll up oursleeves and work together.
Jeremy Julian (34:21):
That sounds amazing.
I haven't gotta figure out howto do a dumpling party.
Maria Zhang (34:25):
But, but if it's just quick and easy, of course
love to order some deliciouspizza and throw some salad.
Everybody loves that.
Yeah.
So, like I said, I'm a mature so
Jeremy Julian (34:35):
I love that.
Maria, I said it earlier.
I ca I am blown away with howthe speed of technology, and
quite frankly, even the lastfive years in my life, the
technology and the advancementhas accelerated so greatly and
it's ultimately when done right,it's gonna make restaurants and
consumers' life so much easier.
So thank you guys for doingthat.
(34:55):
To our listeners guys, you guyshave got lots of choices, so
thank you guys for hanging out.
If you haven't already done so,please subscribe to the show on
your favorite listener and makeit a great day.
Maria Zhang (35:04):
Thank you so.
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
On Purpose with Jay Shetty
I’m Jay Shetty host of On Purpose the worlds #1 Mental Health podcast and I’m so grateful you found us. I started this podcast 5 years ago to invite you into conversations and workshops that are designed to help make you happier, healthier and more healed. I believe that when you (yes you) feel seen, heard and understood you’re able to deal with relationship struggles, work challenges and life’s ups and downs with more ease and grace. I interview experts, celebrities, thought leaders and athletes so that we can grow our mindset, build better habits and uncover a side of them we’ve never seen before. New episodes every Monday and Friday. Your support means the world to me and I don’t take it for granted — click the follow button and leave a review to help us spread the love with On Purpose. I can’t wait for you to listen to your first or 500th episode!