BrokerIQ Roadmap

Status of the build (all milestones through s10 complete and committed).

Done

s0 Scaffold — uv project, pyproject, gitignore, README stub
s1 Core package — config, logging, state/models, hello-world LangGraph agent
s2 GitHub repo pushed (public)
s3 Tools — web search, NAICS lookup, compliance RAG tool
s4 Full graph — supervisor + research/qualification/compliance/report agents, HITL interrupt
s5 RAG pipeline — Qdrant hybrid (BM42 + dense + RRF), cross-encoder rerank, Redis semantic cache
s6 Memory extraction loop on PostgresStore
s7 FastAPI SSE streaming + HITL resume API
s8 MCP 2.0 server exposing compliance_search
s9 Evals harness (offline FakeLLM + live LLM tiers) + promptfoo CI config
s10 GitHub Actions CI + Docker compose (qdrant / redis / postgres / api)
s11 Production docs — README, architecture, API contract, roadmap, final review

Auth on the API (API keys / gateway) before real deployment
Ingestion CLI + scheduler for the carrier corpus (watch a docs directory, re-index)
More corpus: state-specific commercial auto, E&O, umbrella, workers-comp by class
Live LLM-judge evals wired into CI (needs a provider key in GitHub secrets)
Deployment: managed Postgres + Qdrant Cloud + Redis, uvicorn workers behind a load balancer, brokeriq-mcp as a hosted SSE/streamable-http MCP endpoint
Feedback loop: real broker decisions on HITL reviews → retrain ICP thresholds