Enterprise Knowledge Hub — Agentic RAG (Company FAQ / Policy Assistant)

A demo Agentic RAG system that answers employee queries by retrieving from multiple internal sources (PDFs, Confluence exports, runbooks) and external references. The system demonstrates "agency" — an LLM-driven orchestration of Router → Retriever → Verifier → Synthesizer agents — producing grounded answers with provenance and inline citations.

Features

• Multi-source retrieval: PDFs, DOCX, Confluence HTML, structured tables.
• Agentic orchestration:
  • Router Agent chooses which knowledge sources to query.
  • Retriever Agents perform hybrid retrieval (sparse + dense).
  • Refinement Agent rephrases queries if coverage is low.
  • Synthesizer Agent composes answers with inline citations.
  • Verifier Agent computes confidence and detects contradictions.
• Streamlit demo UI showing retrieval path and highlighted sources.
• Test suite for ingestion, retrieval, and end-to-end QA.

Architecture

See docs/ARCHITECTURE.md for diagrams and component breakdown:

• Ingestion → Chunking → Embedding → Vector DB
• Router → Retriever(s) → Aggregator → Verifier → Synthesizer
• Streamlit frontend → Backend API

Tech stack

• LLM: OpenAI
• Embeddings: OpenAI / local embedding model
• Vector DB: Chroma (demo) / Faiss / Milvus (production)
• Backend: FastAPI
• Frontend: Streamlit
• Orchestration: Python (LangChain or custom agent flow)
• Tests: pytest, tox
• CI: GitHub Actions (run tests, lint, build docker)

Quickstart (local demo)

Prereqs: Python 3.10+, pip, Git, (optionally) OpenAI API key

1. Clone

git clone https://github.com/<your-repo>/enterprise-knowledge-hub.git
cd enterprise-knowledge-hub