Description
Summary We need an experienced AI engineer to build a multi-agent research automation pipeline that can take a query, decide whether it actually needs research or can be answered directly, plan out what to look into, gather information from multiple sources, and produce a clean written report — with a built-in quality check that catches thin or incomplete results and loops back to fix them before anything reaches us. This isn't a simple "LLM + search API" wrapper. We're looking for someone who understands proper agent orchestration — routing logic, planning/re-planning, multi-source retrieval, and self-correction loops — not just prompt chaining. What we need: Multi-agent system built in LangGraph (or comparable framework — CrewAI/AutoGen open to discussion) Router/classifier step to avoid unnecessary research on simple queries Planner + Researcher + Writer stages A quality-check/self-correction step with a real feedback loop, not just a linear pipeline Integration with web/news search and a document store (vector DB) for retrieval Clean, modular code — we may extend this ourselves later, so architecture matters as much as function Working prototype we can run and test, not just a design doc Nice to have: Experience with LangSmith or similar tracing/observability tools Experience productionizing agent systems (not just notebooks) Portfolio examples of similar multi-agent or RAG pipelines