How dev-nikhil02/agentic-rag-demo works
Project Type & Purpose
• A Retrieval-Augmented Generation (RAG) pipeline for querying CSV (“books/GOOGL…”), PDF (“financial_reports/…”), and other text sources.
• Provides a simple Python app (app.py) and scripts (rag_pipeline.py) to ingest docs, build embeddings in ChromaDB, and answer user queries via an LLM.
Main Components
• Data Sources: CSV files, PDF reports, notebooks (prototype_Rag_agent.ipynb)
• Ingestion & Preprocessing: rag_pipeline.py reads, splits, and embeds documents
• Vector Store: ChromaDB directory with binary and SQLite files
• Retriever: LangChain-style vector search against ChromaDB
• LLM Interface: OpenAI (or similar) calls wrapped in app.py
• UI/API Layer: app.py exposes a CLI or HTTP endpoint
• Logging: llm_logs.txt for tracking queries/responses
Component Interactions
• Response flows back to app.py → User
• Logs captured from each LLM call
Architectural Patterns & Principles
• Separation of concerns: ingestion, storage, retrieval, and presentation are distinct modules
• Pipeline pattern: rag_pipeline.py defines a linear ETL-style flow
• Modular storage: ChromaDB abstracts vector persistence