I tried to describe the entire process and save it in md files: chat*.md and others. Start with them.
Two independent solutions for the Agentic RAG Challenge — a competition to build the best Retrieval-Augmented Generation pipeline for answering questions over legal PDF documents.
Both solutions were developed with AI coding assistants and explore different approaches to PDF parsing, hybrid retrieval, and answer generation.
| Solution | AI Assistant | Approach | Final Version |
|---|---|---|---|
| rag_challenge_codex_plus | Codex | PyMuPDF + Hybrid RRF + systematic experiments (v1–v16) | v16 |
| rag_challenge_antigravity_free | Antigravity (free) | Docling + Hybrid + LLM Reranking (v0–v22) | v22 |
- Framework: LlamaIndex
- Embeddings: OpenAI text-embedding-3-large (via OpenRouter)
- LLM: GPT-4o-mini (via OpenRouter)
- Retrieval: Hybrid (BM25 + Vector search)
See each solution's README for detailed setup and usage instructions.
AKT team.