09 Current

Beginning of December.

Folder: llm_code

  • now: most core moving parts of a RAG system

  • remain: improve structure + add components for best practices

Foundation

  • config system

Data side

  • ingestion process

  • vector index

  • retriever

Reasoning side - agent

  • LLM

  • tools / utils

  • middleware

  • messages

UI & observability

  • CLI

  • LangSmith tracing

Sam

more details agent.py is the "driver" that calls other scripts.

  • tools

    • rebuild_index()

    • search_docs()

  • middleware

    • trim_history()

    • max_context_chars()

  • CLI

    • index()

    • ask()

Sam

next steps

  • middleware: more to add

  • evaluation: LangSmith Datasets and Evaluation, subgraphs, guardrails

  • graph: add router/branching logic (currently graph is linear)

  • tools: add semantic routing

  • loading: chunk / splitter module (currently default loader)

  • async pipeline: currently 1 step at a time

  • output: want json, currently plain text

  • chat UI

  • .py files: simplify, continue refactoring to match langchain