Milestones

List view

• Phase 4: hybrid RAG, reranker & retrieval eval

  Improve retrieval quality: hybrid dense+lexical search (RRF), CPU cross-encoder reranker, and a downstream retrieval-mode eval. Fine-tuning deferred.

  No due date

  •4/4 issues closed
  100% complete0 open4 closed

• Phase 3: LangGraph, MCP & streaming

  Agentic orchestration + interop: LangGraph application-assistant (HITL), MCP server, end-to-end streaming, MCP client.

  No due date

  •5/5 issues closed
  100% complete0 open5 closed

• Phase 2: Safety, guardrails & eval gating

  Production hardening: API rate-limit + cost ceiling, PII handling, LLM I/O guardrails, two-tier eval gating.

  No due date

  •5/5 issues closed
  100% complete0 open5 closed

• Phase 1: Real data + LLMOps backbone

  Real job ingestion (Adzuna + fallback), Langfuse tracing of the agent, and a Ragas+pytest eval harness. Closes critique gaps F, B, A(starter).

  No due date

  •4/4 issues closed
  100% complete0 open4 closed