ai-customer-support-agent

AI Customer Support Assistant is an enterprise-oriented AI customer support assistant built with Python, LangGraph, FastAPI, React, PostgreSQL, RAG, and Large Language Models (LLMs). It combines LLM-powered intent classification, controlled business tool execution, retrieval-augmented generation, persistent database operations, and human-in-the-loop workflows into a modular customer support architecture. The project evolved from a V1 Streamlit application into a multi-layer V2 architecture with agent orchestration, backend APIs, a customer interface, and a company operations portal.

The system uses LLM-powered intent classification, Pydantic structured intent extraction, confidence-based routing, request validation, and order ID extraction. Business tools include order status lookup, delivery status lookup, refund status lookup, order cancellation, and human support escalation. General customer questions are routed to a RAG pipeline using knowledge-base ingestion, text chunking, Sentence Transformer embeddings, FAISS vector similarity search, and context retrieval. LangGraph orchestrates conditional routing, tool execution, error handling, human-in-the-loop interrupts, approval/rejection/escalation branches, and final response generation.

High-impact or ambiguous requests can be paused for human review. The system provides persistent human review requests, a company-side review queue, approve/reject/escalate actions, backend state persistence, and customer-side status updates. PostgreSQL stores orders, refunds, support tickets, and human review requests, with database access isolated through a repository layer.