Sunday Coffee & Code: Agent 6 is online, RAG improvements are on their way.

Less time available this week than usual but still have progress. ● Agent 6: Finished up Agent 6 that takes the initial response and the QA input and creates an updated version. Used the orchestrated set of agents this week to respond to an RFP - completed in under 2hrs and submitted (will be interesting to see the results). ● High Risk/High Friction: In the process of preparing the final version I realised there is room for another agent to prepare an even more polished version and also that this is as far as I would likely take it. An RFP or Proposal is on the high risk/high friction end of the spectrum and human input is definitely required prior to submission. ● RAG: Improvements are underway. LightRAG is installed and running, creating data stores and knowledge graphs (snippet below). Couple of issues to work on - Server UI isn’t replying using knowledge in the knowledge base (pretty fundamental - not sure if this is just the GUI version or not - will test programmatically) - There may be a bug for embedding models that don’t use 1024 dimensions (to be confirmed - https://lnkd.in/ghMdvEjQ). - Ollama doesn’t seem to support re-rankers - will have to look at other solutions for this (switch to vLLM or maybe just go all-in on hosted models). All in all some good progress and a lot learned. Next on the roadmap: - Further testing on LightRAG and re-ranker decisions. - Agent 007?