RAG Application (Agentic Follow-up QA with Page Citations)

An advanced, SOLID-architected Retrieval Augmented Generation (RAG) app that:

• Answers questions strictly based on provided PDF documents
• Includes page numbers for each cited source
• Supports follow-up questions via query rewriting
• Compatible with OpenAI and Ollama models (switchable via config or CLI)
• Shows a progress bar for all ingestion and query steps
• Comes with unit tests and mock providers for offline testing

Features

• PDF ingestion with page-aware chunking
• Embedding-based retrieval (OpenAI or Ollama; Ollama fallback to OpenAI/Mock if embeddings API not available)
• LLM-based query rewriting and relevance grading (Ollama chat falls back to /api/generate when /api/chat is missing)
• Answer generation constrained to retrieved content with citations
• Clean provider interfaces (SOLID) and dependency inversion
• CLI for interactive Q&A over your PDFs

Quick Start

1. Install dependencies with python 3.10

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

2. Configure model provider

• OpenAI: set environment variable OPENAI_API_KEY.
• Ollama: ensure ollama is running locally (default http://localhost:11434). For embeddings support you need a recent Ollama and an embedding model, e.g.:
  • ollama pull nomic-embed-text
  • If /api/embeddings is not available, the app will fall back to OpenAI (if OPENAI_API_KEY set) or a Mock embedder.

You can configure via .env (optional):

OPENAI_API_KEY=sk-...
OLLAMA_BASE_URL=http://localhost:11434
OPENAI_MODEL=gpt-4o-mini
OPENAI_EMBEDDING_MODEL=text-embedding-3-small
OLLAMA_MODEL=llama3.1
OLLAMA_EMBEDDING_MODEL=nomic-embed-text

3. Run the CLI

python -m rag_app.main --pdf docs/your.pdf --provider openai
# or
python -m rag_app.main --pdf docs/your.pdf --provider ollama

Then type your questions. Follow-ups are supported. Answers include page citations.

Tests

Run tests with:

pytest -q

Structure

• rag_app/config.py – configuration and settings
• rag_app/interfaces.py – provider interfaces and core contracts
• rag_app/pdf_loader.py – PDF page extraction
• rag_app/text_splitter.py – page-aware chunking
• rag_app/embeddings.py – embedding providers (OpenAI, Ollama, Mock)
• rag_app/vector_store.py – simple cosine similarity index
• rag_app/llm.py – chat LLM providers, query rewriter, grader, answerer
• rag_app/agent.py – orchestrates the end-to-end RAG flow with progress bars
• rag_app/main.py – CLI entry point
• tests/ – unit tests for splitter, store, and agent