Quality Assurance
RAG.md · 8 documents
RAG Deep Dive Part 7: Evaluation and Debugging RAG Systems
**Series:** RAG (Retrieval-Augmented Generation) A Developer's Deep Dive from Scratch to Production
RAG Evaluation
title: RAG Evaluation
embedding-first-chunking-second-smarter-rag-retrieval-with-max-min-semantic-chunking
id: embedding-first-chunking-second-smarter-rag-retrieval-with-max-min-semantic-chunking.md
Fix Summary: Matroska Adaptive Chunking for 128D Embeddings
**Critical Bug**: [clustering_rpn.py:43-48](knowledge3d/cranium/clustering_rpn.py#L43-L48) was truncating 128-dimensional embeddings to **4 dimensions**:
SKILL-FORGE AUDIT: sop-dogfooding-pattern-retrieval
**Audit Date**: 2025-11-02
Week 2 Requirements - Completion Status
**Project:** ChatBot Application with LLM and RAG Integration
MCP Fact-Check Design Document (Enhanced)
1. [Overview & Objectives](#overview--objectives)
JudgeIt (From SuperKnowa)- Automatic Eval Framework for Gen AI Pipelines
The single biggest challenge in scaling any GenAI solution (such as RAG, multi-turn conversations, or query rewriting) from PoC to production is the last-mile problem of evaluation. Statistical metrics (like BLEU, ROUGE, or METEOR) have proven ineffective at accurately judging the quality of AI generated text, leaving human evaluation as the only reliable option for Enterprises. However, human evaluation is slow and expensive, making it impossible to scale quickly. This is where 'JudgeIt' comes