How syedasad-kiwi/agentic-rag-poc works
Project Type and Purpose
• This is a proof-of-concept, full-stack Retrieval-Augmented Generation (RAG) system.
• It ingests raw PDF/DOCX documents, chunks and embeds them, stores embeddings in PostgreSQL+pgvector, orchestrates multi-agent workflows with CrewAI and Ollama, exposes an OpenAI-compatible FastAPI, and provides a web interface via OpenWebUI.
File Structure Analysis
• Top level
• Dockerfile, docker-compose.yml: container orchestration
• api.py, main.py: API server and CLI entry points
• init-db.sql: database schema setup
• data/raw/: source documents
• requirements.txt / requirements-docker.txt: dependencies
• src/
• config/settings.py: environment and configuration
• data_ingestion/ingest.py: document parsing, chunking, embedding
• evaluation/: test datasets and evaluation scripts
• rag_system/
– agents.py: CrewAI agent definitions
– crew.py: multi-agent orchestration
– tools*.py: custom retrieval/generation utilities
• Tests: test_tool.py
README Insights
• Technologies: Docker, Docker Compose, Ollama, LlamaIndex (DoclingReader), FastAPI, PostgreSQL+pgvector, OpenWebUI