The problem

Support teams answer the same types of questions repeatedly: “What’s my order status?”, “What’s in the knowledge base?”, “Summarize ticket trends.” Doing this with a single generic chatbot is brittle; different questions need different backends—RAG over docs, SQL over DB, or analysis over aggregated data.

The solution

Customer Support AI uses LangGraph to orchestrate three agents: RAG (Pinecone + LLM for knowledge-base answers), SQL (natural language to read-only SQL over PostgreSQL), and Analyzer (data analysis and insights). An orchestrator routes each query to the right agent(s) and returns a unified response.

Without specialized agents

One-size-fits-all chatbot; no clean separation between docs, DB, and analysis—hard to scale and secure.

With Customer Support AI

RAG + SQL + Analyzer agents with automatic routing; security (API key, read-only SQL, CORS); Docker-ready.

What it does

  • RAG agent – Retrieves context from Pinecone, generates answers from knowledge base.
  • SQL agent – Translates natural language to SQL, runs read-only queries on PostgreSQL with safety checks.
  • Analyzer agent – Analyzes data and returns insights (e.g. trends, summaries).
  • Orchestrator – Routes queries to the appropriate agent(s) automatically.
  • Security – API key auth, input validation, SQL injection prevention, CORS, non-root Docker.

Tech stack

Backend: FastAPI, LangGraph, Pinecone, SQLAlchemy, PostgreSQL. Frontend: React, Vite, Axios. Docker Compose for full stack.

Next steps

Roadmap & ideas

  • Demo (TBD) – hosted or self-hosted with sample data.
  • More agents (e.g. ticket creation, escalation) and routing rules.
  • Auth and multi-tenant support for production use.