Building with Neo4j Relate: NorthWindPLUS Use Case

Zusammenfassung

The Neo4j community is showcasing an innovative implementation of the NorthWindPLUS dataset using Neo4j Relate. This project demonstrates how to leverage Relate for offline testing, vector search, and multi-interface business logic, including automated scripts, AI agents, and React UIs. It highlights the benefits of zero-setup graph development while providing valuable feedback on current tool limitations and workflow improvements for developers.

embergarden-51 OP

Sharing an interesting use case for Relate: https://github.com/Flyer-Boy/NewApproach/tree/main/NorthWindPLUS/Neo4jRelate - I hope you like it.

happybeacon-37

wow this is super cool! i cant believe that you already build this with relate, how do u like relate so far, any feedback for the team

happybeacon-37

what was smoother with relate than doing the same thing in regular Neo4j, and what was rougher?

happybeacon-37

sme inspirations if you are still thinking about what to build with relate

@happybeacon-37 what was smoother with relate than doing the same thing in regular Neo4j, and what was rougher?
embergarden-51 OP

Thanks! Loving it so far.

Smoother: zero setup (pip, one file, same notebook on Windows and Colab), the same Cypher as Aura (the simulation's vetting Cypher ran in Relate unchanged, minus APOC), Relate as a test double for Aura in offline tests (very cool!!), full-text + vector search in one statement with indexes updated on write, and relate pull for scratch copies.

Rougher: one writing connection per store even in one process ("storage locked" in notebooks), no APOC (I strip apoc.lock lines. And I am aware that my use case is not the common one you might get. I use the Graph for transacting.), full-text without stemming ("restocking" ≠ "Restock"), pull only in the CLI and not in the Python API, Temporal values needing to_native() to go back in as parameters, and push never deleting (my state changes delete relationships, so I commit through governed Cypher instead).

embergarden-51 OP

By the way, I just pushed a new version, 2.3.3, that lets the agent submit POs (one of the simulation loops), so you can do this via the Agent instead of the Python script.

embergarden-51 OP

Stay tuned as I am now building a third front end to the model. A traditional one with a React UI. In this way I will have three different front ends performing the same business function (PO creation) - one running as an automated script (python), an AI Agent (conversational UI) and the traditional web UI.

@embergarden-51 Thanks! Loving it so far. Smoother: zero setup (pip, one file, same notebook on Windows and Colab),…
happybeacon-37

this is great, just forward it to the internal team in case this missed it

@embergarden-51 Stay tuned as I am now building a third front end to the model. A traditional one with a React UI.…
happybeacon-37

oh thats cool, curious which one ends up being the most reliable for PO creation

embergarden-51 OP

Look, both are equally reliable as both are built on the same foundations and share most of the code, cypher queries, etc. I personally like the UI better than the conversacional one. But sometimes the conversational could be more patriarcal if you are not in front of a computer. Let's say that you are at the airport boarding a flight and need to re-submit a PO urgently.

embergarden-51 OP

In my latest update, I introduced a "Stuck Order" functionality. Neo4j Relate pulls a subset of the Aura graph to simulate the business forward, identifying the products and quantities that need to be ordered from suppliers (Purchase Orders) to unblock customer orders that cannot be fulfilled due to insufficient inventory. Check it out.

embergarden-51 OP

By the way, this is a TypeScript + React UI, using an AI Agent using Neo4j Relate to make this possible.

Das Neueste von Neo4j