Graph RAG integration update

make this short: I got it to work and now it more accurately recalls the information I ingest and recall using graph RAG with a reasoning agent using evaluations. What is more is that it is fairly quick and is not hallucinating. That is the best part is that it seems to dramatically help reduce that. It could be just me getting the graph functionality to finally work as I wanted, but it is also the integration of the evaluations. I still have a lot of work to do on it and a lot more functionality to add, but as it is now I can perfectly and easily chat with my reddit history or any other set of documents I want. I will probably just use this instead of NotebookLM once I have it completed. It works locally and the data never leaves your device which is perfect for a lot fo more sensitive use cases such as hospitals and such. Next I need to add Auth and then it would be something you could ship. Then just maintain it and I imagine that is what the AI Developer role is at companies that they are hiring for. That is the job role my blog is meant to help work towards, basically just finding real world applications which can be solved with AI and implementing it with software engineering. More like a developer role who fully integrates AI into what they create.

Here’s a much shorter version with the same meaning: I finally got it working: the graph RAG with a reasoning agent now recalls information accurately, runs fast, and barely hallucinates. Some of that is from finally getting the graph right, but the evaluation layer also helps a lot. There’s still plenty to build, but as it stands I can seamlessly chat with my entire Reddit history—or any document set—locally, with zero data leaving the device. Once I add auth, it’ll be shippable. Maintaining systems like this is basically what an AI Developer does: find real problems, apply AI, and integrate it cleanly into real software. That’s exactly what my blog is trying to teach.