dalgona039/TreeRAG
TreeRAG: Hierarchical RAG platform that indexes dense PDFs into logical trees for precise retrieval, 100% traceability, and LLM-guided deep traversal. Production-ready with enhanced security and streaming indexing.
What's novel
TreeRAG: Hierarchical RAG platform that indexes dense PDFs into logical trees for precise retrieval, 100% traceability, and LLM-guided deep traversal. Production-ready with enhanced security and streaming indexing.
Code Analysis
9 files read · 3 roundsA hierarchical RAG system that indexes PDFs into tree-structured JSON, traverses relevant branches using DFS or beam search with LLM-based relevance scoring, and generates grounded answers with citations, hallucination detection, and contextual compression.
Strengths
Genuine algorithmic depth with beam search, TF-IDF scoring, similarity-based context compression, and multi-signal hallucination detection. Sophisticated error handling with retry logic, rate limiting, and adaptive threshold adjustment for over-filtering recovery.
Weaknesses
Heavy dependency on Gemini API for core traversal logic (LLM calls at every node evaluation), making it expensive and slow. The error recovery and adaptive policy modules couldn't be fully verified, and the system's performance characteristics under load are unclear.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
88
Contributors
1
Files
184
Active weeks
7
Repository
Language
Python
Stars
2
Forks
0
License
MIT