djleamen/doc-reader
Large document Q&A agent using RAG
What's novel
Large document Q&A agent using RAG
Code Analysis
9 files read · 3 roundsA Django-based RAG document Q&A system that ingests PDF/DOCX/TXT/MD files, chunks and embeds them into FAISS vector indexes, retrieves relevant context for questions, generates LLM answers, and validates semantic coherence across the retrieval-to-generation pipeline with automatic fallback actions.
Strengths
The semantic coherence validation layer is a genuinely thoughtful addition that makes RAG failure modes visible rather than silent, and the Django views show production-grade thinking with thread-safe LRU caching, session isolation, per-file error isolation in batch uploads, and careful input validation. The codebase is well-documented with consistent docstrings, clear naming, and clean separation between the core RAG engine, vector store abstraction, document processing, and web/CLI interfaces.
Weaknesses
The coherence validator hardcodes a 1536-dimension zero-vector fallback for embedding failures, which would silently produce meaningless similarity scores if a different embedding model is configured. The test suite exists (24KB in rag_app/tests.py plus tests/test_utils.py) but I could not verify its depth, and the Azure pipeline (described as experimental) could not be verified as functional. The core RAG pipeline is a fairly standard LangChain+FAISS composition, so the novelty is concentrated
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
145
Contributors
4
Files
50
Active weeks
21
Repository
Language
Python
Stars
2
Forks
1
License
MIT