Nexus HR Agent
An AI-powered internal operations assistant built with AWS Bedrock, RAG, and FastAPI. Ask questions about company documentation or create support tickets — the agent routes your request to the right skill automatically. Built to learn and mirror production RAG + agentic patterns, alongside studying a production Strands Agents + S3 Vectors system at work to understand every architectural decision by building a working equivalent.
Product demo — the chat agent handling documentation Q&A and ticket creation
How I validated it — live eval suite, 12 test cases scored against the RAG pipeline
- RAG pipeline over fictional company documentation: LangChain chunking, Amazon Titan embeddings via Bedrock, ChromaDB as the vector store — full document ingestion and semantic retrieval
- Agent with tool calling: two-round-trip Bedrock format (initial call → tool invocation → second call with results). Two tools: search_docs (RAG retrieval) and create_ticket (support ticket creation)
- Intent-based skill routing (documentation vs. support), stateful multi-turn conversations with session persistence, and conversation summarization for long sessions
- FastAPI backend; Next.js + React + TypeScript + Tailwind chat frontend
- FastMCP stdio server — exposes the agent as an MCP-compatible tool for use with Claude Code or any MCP client
- LLM-as-judge evaluation pipeline for measuring retrieval and response quality
PythonFastAPINext.jsReactAWS BedrockChromaDBLangChainMCP
View code →