balavenkatesh3322/rag-doctor
🩺 Agentic RAG pipeline failure diagnosis tool. Tells you why your RAG failed — chunk fragmentation, retrieval miss, position bias, hallucination, or query mismatch — with a root cause ID and concrete fix. CLI + Python SDK + Ollama support.
What's novel
🩺 Agentic RAG pipeline failure diagnosis tool. Tells you why your RAG failed — chunk fragmentation, retrieval miss, position bias, hallucination, or query mismatch — with a root cause ID and concrete fix. CLI + Python SDK + Ollama support.
Code Analysis
17 files read · 4 roundsDiagnoses why a RAG pipeline produced a wrong answer by running six structural and semantic checks (retrieval recall, position bias, chunk quality, hallucination grounding, query rewriting, chunk optimization) in a priority-ordered decision tree and returning a classified root cause with an actionab
Strengths
Clean tool-based architecture with a practical 4-tier embedding fallback chain (sentence-transformers → Ollama → TF-IDF → char-frequency) that works with zero hard dependencies; the hallucination tracer's claim-level grounding and the chunk optimizer's grid search with recall@k measurement are genuinely useful algorithms.
Weaknesses
The 'agentic' branding overstates what is a sequential if/else decision tree; models.py contains dead duplicate classes (Document, ToolResult) that are never imported, and the hallucination tracer's hardcoded 0.80 pass threshold is inconsistent with its configurable 0.60 faithfulness_threshold parameter.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
6
Contributors
0
Files
79
Active weeks
1
Repository
Language
Python
Stars
1
Forks
1
License
MIT