Ep42: RAG and Java + AI-generated content and MCP

Star Wars Day is nearly here, and this episode is stacked with tech goodness to celebrate! I’m diving into highlights from the Neo4j ecosystem, starting with an early look at the Using Neo4j with Java course—perfect for getting started with the Java driver in a framework-less setup.

Also in this episode:

• ⚙️ Behind-the-scenes of APOC + Pinecone integration

• ✨ Part 2 of my Intro to Retrieval Augmented Generation series

• 🎥 My recent guest spot on Neo4j Live, discussing the Developer’s Guide to Building a Knowledge Graph

• 🤖 A fascinating series on an AI content experiment from Mark Heckler

• 📚 Michael Hunger’s must-read blog on the Model Context Protocol (MCP)

May the Fourth be with you!

Episode Transcript

(00:05):
You are listening to the BreaktimeTech Talks podcast, a bite-sized tech
podcast for busy developers where we'llbriefly cover technical topics, new
snippets, and more in short time blocks.
I'm your host, Jennifer Reif, anavid developer and problem solver
with special interest in data,learning, and all things technology.

(00:26):
Star Wars Day or May the Fourth isapproaching this weekend, and I'm
already definitely ready to geek out.
There is so much content that hasbeen published recently, so I'll
try to hit a few highlights thatI've dug in on just this week.
But I have a ton of content tabs openthat I still have yet to catch up on,
so that'll probably percolate overthe next several weeks' episodes.

(00:46):
Plus I've produced some content aswell, so this will be an episode
that is jam packed with tidbitsand threads for you to pull on.
The first thing that I want to bring upthis week is there will be a new Neo4j
Graph Academy Java application course.
I hinted at this a bit last weekand said I would go into a bit
more detail this week on it.
It's gonna be calledUsing Neo4j with Java.

(01:09):
Hopefully it will be releasedin the next week or so.
We'll kind of keep an eye on that.
It should show as a coming soon courseon Neo4j Graph Academy right now, but
we'll see when we can get that live.
It's an upcoming Java driver course.
The goal for the course is just toshow you the foundations of working
with Neo4j and the Java driver.

(01:31):
How to connect to Neo4j from Java,how to run queries, and handle
the results that are coming back.
Then, once you would finish thiscourse, you could go on to the full Java
application course that would follow.
That does need revamped right now.
It is also on my to-do list, sohopefully I will have some updates
to that in the next few weeks.
But for the time being, thisis a great start, just the

(01:54):
Using Neo4j with Java course.
It is a framework-less app, soagain, it's just using Neo4j
and the vanilla Java driver.
No framework involved.
Shows you how to spin up theconnection details, run queries,
and so on, as I mentioned.
This was a lot of fun.
I worked with this course with anothercolleague of mine and we collaborated on

(02:16):
this effort, and through that I learneda little bit more about the Java data
type and Cypher data type mappings forthings like temporal types, spatial
types, and some error handling as well.
I really enjoyed diggingjust a little bit deeper.
As I've done development over the last fewyears, I've focused mostly on the Spring

(02:37):
ecosystem, which has been pretty cool.
But there are a lot of thingsthat Spring does out of the box.
So I really thought this was neat tolook specifically at just the plain Java
driver and explore some very Java centricthings and maybe help me understand
some Spring things that they're doingunder the hood, as well as maybe explore
possibilities with other applicationsand frameworks and so on down the road.

(03:01):
I really enjoyed this.
Learned a lot more about how Neo4j andJava interact with one another, how
you would use just the Java driver forconnecting to a database such as Neo4j.
So I'm really excited to see thiscome out in the next week or so.
keep an eye out for that.
It is a very short form course.
I think it's only gonna be about anhour, so very easy to get up and running.

(03:22):
If you just wanna look at an introductionfor working with Java and Neo4j,
this would be a great starting point.
The next thing I wanted to talk aboutis I mentioned a few weeks ago that
I had played around with Neo4j'sAPOC (Awesome Procedures On Cypher).
It's a utility library for Neo4j thatextends Neo4j's functionality, but APOC
provides some procedures for connectingto Pinecone, which is a vector database.

(03:46):
I had said I'd had some troubletrying to figure out how to
connect to Pinecone using APOCand run some of the functionality
that should be available there.
And so I pulled down the repository afew days ago and started figuring out
how I could create a pull request forupdating the functionality for some
of the API changes that Pinecone hasmade in the last couple of months.

(04:08):
And started playing around withit and actually realized when I
dug into the code for APOC thatthere's not functionality missing.
It's actually all there.
But the configuration was kind of hidden.
Just like it took me a while to figureout how to format the header's key
configuration for specifying the API Keyas a request header, you can actually add

(04:30):
all the required fields for Pinecone'sindex spec in the configuration as well.
Again, the documentation doesn'tmake this super clear, and right
now it says something like optionalconfig, which actually, depending
on the Pinecone procedure you'reusing, it's not optional, right?
There are certain things that are requireddepending on the procedure and Pinecone,

(04:50):
at least right now that I can find.
I haven't been able to getthe host key first parameter.
working by using the Pinecone host name,but I just end up leaving that null and
then specifying the API key in the header.
And of course, right now when youdo that, you have to specify the API
key so it knows where to connect.
And so that is a requiredconfig at this point.

(05:11):
Same thing would be if you'retrying to create a new index.
You need to specify, first of all,whether it's a serverless index or a
pod index, and then some of the detailsthat go along with each one of those.
Again, that would not be optional aswell, depending on what you're doing.
It's just, it wasn't clear howto specify some of the config.
And I am unable to get that hostkey as the first param, but I may

(05:33):
just be doing something wrong.
According to docs and other peopleI've talked to, that host Pinecone URL
should work as the first parameter,and for some reason it's not for me.
I'm hoping I'm just doing somethingwrong, and I'll figure that out soon.
I will hopefully submit a PR thoughon some of the documentation and
examples maybe in the near future.
Then I have some content as wellthat I got released this week.

(05:55):
The first was, I had submitted lastweek an Intro to Retrieval Augmented
Generation, part one, talkingabout a few things, a few concepts
of the Generative AI landscape.
This week, I published part two, whichgoes and stacks a few more concepts on
top of what we talked about in part one.
And I talk a little bit aboutgenerative AI as layers.

(06:16):
I've started thinking of Gen AI aslayers, and I kind of like that analogy.
We talk about vector rag, thengraph rag, agents, and then MCP.
I talk about what each is and whenthey're beneficial, when you should
use one over the other and so on.
Just to kind of get you started.
Again, covering a lot of thesefoundational concepts and then providing
external links where you can go tolearn more information or build some of

(06:39):
that knowledge on your own elsewhere.
Then I also had a guest appearanceon Neo4j Live, which is Neo4j's live
stream that occurs every so often.
And I talked about my ebook thatI released late last year, which

is called Developer's Guide: How to Build a Knowledge Graph. (06:52):
undefined
And that is available as a freedownloadable resource, but I talked a
little bit about it on the live stream.
I walk through the knowledge graphebook in this session, a little
bit about the tools, how to builda graph and a knowledge graph,
what the differences are there.
Show the repository links forwhere to pull in and import data.

(07:15):
There's also a couple of screens thathave changed in the Aura Console for
Neo4j that show up differently thanthe screenshots shown in the book.
So I talk a little bit about the changesthat have been made in the UI, as well
as the tools, and tips and tricks, andanswer some live questions as well.
So if you're interested in that,I'll link the YouTube video as
well as the downloadable ebook.

(07:35):
Now, we can spin on tothe content section.
Again, I have tons of tabs ofcontent open, but these were the
two things that I focused on readingand catching up on this week.
The first is a blog series produced byMark Heckler on AI generated content.
What happened here is there are tworepositories, two code repositories.

(07:57):
One is a blog generator AI, andthe other is a blog editor AI.
There's some resources and promptsoutlined in each one of these, but
the blog generator takes a promptinput, as well as a particular topic
and some guidelines and requirementsfor the content it needs to produce.
And then it produces a piece of content,a short blog on that topic and sends

(08:21):
it over to the editor service, theother code repository, if you will.
And the blog editor AI has somerequirements and a prompt and resources
that have been outlined by the developerand then takes the blog generated
content and edits it or analyzes itand approves or rejects the content,
and then submits some suggestions back.

(08:43):
And there's a few iterations thathappen there, replicating a writer
editor that we use in the human worldwhere somebody would write a piece
of content, then an editor wouldreview it and send back comments.
Then the writer would take thoseand create a revision and send that
back to the editor, and the editormight have a few more additions and,
and revisions to be made and so on.
And that's exactly what these ais aredoing in this particular blog series.

(09:08):
The content is published on Mark Heckler'sblog and shows, and very clearly says
this is AI generated content, butshows the output of what the writer
and editor came up with at the end, aswell as whether it was finally approved
or rejected after so many iterations.
So a series of max iterations didn't letit go on for like 20 or 50 revisions.

(09:29):
It was only I think four, threeor four or something like that.
And most of the blog posts alsoinclude some developer notes by Mark
on just his thoughts and opinionsand interpretation of the content
that was produced on the other side.
This I found was a reallyinteresting experiment on AI
generated content and feedback.
So using an AI to generate contentand then using another AI to

(09:52):
evaluate and revise that content, twoseparate AIs working in tandem here.
I thought this was really interesting.
There's I think now 15 blog posts thatare, that are published to that site.
So if you wanna check that out, aswell as the code rep repositories
that go along with that, I will linkto everything in the description.
Then the second piece of content, andthe last one for today, is that in

(10:15):
researching for my Intro to RetrievalAugmented Generation blog post part
two that I did this week, I looked alittle bit at Michael Hunger's blog
post on MCP It's called Everythinga Developer Needs to Know about
the Model Context protocol or MCP.
MCP launched late last fall and tookthe generative AI market and caught it

(10:36):
up in a whirlwind, that is for sure.
But I really loved Michael Hunger's takeon just the overall sphere of what's
going on with MCP, what it is, how tothink of it, the benefits of it, why it
has kind of taken the world by storm.
It's a new standardization forworking with AI applications and LLMs.

(10:58):
Shows some examples and diagramsof architectures, some examples
of servers and resources, plussome MCP integrations that Michael
and the team have built as well.
And then it also talks about someof the limitations or considerations
that you need to think of with MCP asit stands right now in the industry.
Then finally, the blog post wraps upwith plenty of third party resources

(11:20):
for getting started with MCP.
So whether you're looking forvendor perspectives on MCP or
maybe some integration examplesor you want to look at other
people's thoughts and perspectiveson the introduction, what MCP is.
You can look at a fewdifferent options there.
There's a video format, there'ssomebody else explaining MCP and so on.

(11:40):
So lots of options there if you just needa place to kind of formulate your thoughts
on MCP and get an idea for what it is.
I thought this blog post was extremelythorough and very, very helpful.
And formulated the content in my Introto Rag part two blog post as well.
I will link that.
I thought Michael Hunger'spost here on MCP was fantastic.
This episode I gave some details aboutthe soon to be released Using Neo4j

(12:04):
with Java online Graph Academy course,some updates on adventures with using
Neo4j APOC to connect to the PineconeVector database, and published my second
part of the Intro to RAG blog series.
Next, we talked content with aninteresting AI generated and edited blog
series by Mark Heckler and a fantasticoverview of MCP by Michael Hunger.
Thanks for listening to thisweek's fire hose of information,

(12:27):
and May the Fourth be with you.