Multi-Agent Customer Support Platform

Looking for a developer with real hands-on experience building agent systems, not just tutorial-level LangChain work. We're building a multi-agent customer support platform instead of one general-purpose chatbot. Each agent has a specialized job, and everything runs inside a single LangGraph StateGraph. What the agents do: - Router classifies each customer message and directs it to the right specialist - Order agent pulls real order, refund, and shipment data from the database - Policy agent answers questions using retrieval, with self-grading, query rewriting, and web fallback if needed - Troubleshooting agent runs real diagnostic checks instead of generic advice - Quality-check agent reviews every draft reply and escalates low-confidence answers to a human What we're providing upfront: - Seeded database with sample orders, refunds, and shipments - Vector store with policy/product docs already embedded and indexed - Fixed diagnostic functions for the troubleshooting agent - Repo scaffold with configuration and deployment tier set up - Basic frontend shell ready to be connected Core responsibilities: - Build all 5 agents and wire them into one LangGraph graph - Implement short-term and long-term (per-customer) memory - Build a background job for deferred follow-ups (e.g., refund notifications) - Set up an eval dataset with automated scoring for accuracy and groundedness - Deploy the graph and connect it to the provided frontend Tech stack: Python, LangChain, LangGraph, LangSmith, Chroma or Qdrant, web search fallback tool. Requirements: - Comfortable with Python and has used LangChain or LangGraph before - Has built a chatbot or AI agent that connects to a database or documents - Understands how retrieval-based (RAG) systems work - Open to learning tracing/eval tools like LangSmith if not already familiar - Has deployed at least one AI project so it's live and usable, not just running locally - Can show past examples of similar agent or chatbot projects