💽⏏️ Click here to return home

Search

SearchSearch

Neo4j

Jan 01, 1970, 3 min read

Here are the repos that I’ve worked with:

  • chinese dna app that I worked with and combined with michael hungers react-force-graph
  • GRANDstack Citibike Tutorial Dashboard
  • Willow Grandstack app (Building a zillow clone from scratch)
  • LLM knowledgebase

Here are the applications that made by forking/remixing the above repos:

GenAI GraphRAG Applications:

  • LINK [neo4j tutorials and my fork]
  • LINK [docker neo4j snowflake bot?] Overview: Learned the basics the forked the repo to do my own custom project. Project entailed:
    1. extracting unstrutured data from a pdf
    2. semantically/agentically chunking that data.
    3. embedding the data
    4. using prompt engineering templates from lang chain & text2cypher to produce automated queries that align with how I’m seeking to traverse the structure the knowledge graph inside Neo4j in an expensive and high performant way. Defining indexes/schemas that improve overall performance
    5. Lastly, using a chat bot to interface with the document and test output (how accurate the responses were). The document in question was Wild Mushrooms. I acquired this prior to my first foraging experience with a lovely company called “No Taste Like Home” located in Western North Carolina.

What I’ve learned: The use case for RAG and GraphRAG utilizing Neo4j’s UIUX. Long context is performance is relatively strong without the effort setting up custom RAG workflow. RAG still maintains theoretical utility if you have an extremely well defined application and chunking strategy. I think the primary app would be something that is a safe bet to improve ROI rather than a speculative one. Likely, a workflow you are seeking to automate information entry/retrieval for business decisions. Something that RPA sought to solve in the past. Garbage in garbage out is and always will be the name of the game with AI training data. Before taking on a speculative GenAI project it would be important to make sure your processes are in order and paperwork structure doesn’t deviate too much from company standards. Otherwise each new document requires a new chunking strategy.

Here are some videos I watching to dust the cob webbs off:

Fullstack apps:

1st merge back foodhobo to main and then rename to GRANDstack2025 2nd clone and rename foodhobo in a newly initialitzed repo

  • foodhobo (what won me 3rd place in the blockchain competition) GRANDstack
  • citibike snowflake merge and pdf chat bot GRANDstack
  • zillow grandstack?

OR you could create a new repo? nah because you want to be able to showcase the dependecy updates

Incomplete apps/sidequests:

Product Strategy Platform Engineering AIOps see

Related:

Artificial Intelligence Bleeding Edge:

https://hkust-nlp.notion.site/simplerl-reason https://research.google/blog/chain-of-agents-large-language-models-collaborating-on-long-context-tasks/

AI Slop

We define “slop” as:

Content that is mostly-or-completely AI-generated that is passed off as being written by a human, regardless of quality.

https://benjamincongdon.me/blog/2025/01/25/AI-Slop-Suspicion-and-Writing-Back/

Graph View

Backlinks