compliance-aware-rag

An implementation of Retrieval-Augmented Generation (RAG) system with enterprise-grade compliance controls.

Project Philosophy

This project builds:

• A stable LLM baseline
• A proper ingestion pipeline
• Retrieval grounding
• Access control
• Policy enforcement
• PII protection
• Measurable evaluation

System Architecture (Target State)

User
  ↓
FastAPI
  ↓
Authentication Layer (???)
  ↓
Policy Engine (???)
  ↓
Retriever (???)
  ↓
LLM (Ollama)
  ↓
PII Redaction (???)
  ↓
Response

Current Status

Phase 1 — Plain LLM Baseline

Architecture:

User → FastAPI → Ollama → Response

Exit Condition: API consistently returns model responses in < 30 seconds.

Tech Stack

• Python 3.12.3 — Core language
• FastAPI — Web framework
• Ollama — Local LLM inference
• Uvicorn — ASGI server
• Requests — HTTP client

Design principle: Everything runs locally. No external APIs or cloud dependencies.