IdeaCredIdeaCred

djleamen/doc-reader

77

Large document Q&A agent using RAG

What's novel

Large document Q&A agent using RAG

Code Analysis

9 files read · 3 rounds

A 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

Innovation
5 (25%)
Craft
68 (35%)
Traction
16 (15%)
Scope
84 (25%)

Signal breakdown

Innovation

Not Fork+1
Code Novelty+1
Concept Novelty+1

Craft

Ci+5
Tests-4
Polish+0
Releases+3
Has License+5
Code Quality+22
Readme Quality+15
Recent Activity+7
Structure Quality+5
Commit Consistency+5
Has Dependency Mgmt+5

Traction

Forks+6
Stars+6
Hn Points+0
Watchers+0
Early Traction+0
Devto Reactions+0
Community Contribs+4

Scope

Commits+8
Languages+8
Subsystems+13
Bloat Penalty+0
Completeness+7
Contributors+7
Authored Files+12
Readme Code Match+3
Architecture Depth+5
Implementation Depth+8

Evidence

Commits

145

Contributors

4

Files

50

Active weeks

21

TestsCI/CDREADMELicenseContributing

Repository

Language

Python

Stars

2

Forks

1

License

MIT