Home

SmartAssist — Portable RLHF + RAG + MCP Learning System

An AI assistant that remembers every mistake and never repeats them — on any codebase.

Version 4.0 · February 2026 · Joey Rahme

---

What Is SmartAssist?

SmartAssist is a portable, pip-installable Python package that adds a learning layer to Claude Code. It combines three technologies:

Technology	What it does
RLHF	Learns from your feedback — thumbs up/down, corrections, angry signals. Updates reliability scores per category using Bayesian statistics.
RAG	Hybrid semantic search over curated lessons. Converts text to 1024-dim vectors with BAAI/bge-m3, combines vector + BM25 keyword search, then cross-encoder reranks results.
MCP Server	On-demand knowledge retrieval via 3 tools: rag_search, rag_dashboard, and rag_feedback. Claude calls them mid-conversation when relevant.

Key Numbers

Metric	Value
Raw Feedback Events	1,991
Curated Vector Docs	100
Categories	6
Tests Passing	59
Tool Calls Logged	20,070+
Source Files	29
Avg Search Latency	838ms
Search Hit Rate	54%

Quick Start

# Install once globally
pip install -e ~/Github/SmartAssist

# Initialize in any project
cd ~/your-project
smartassist init

# That's it! MCP server and hooks auto-detect the data directory.

Documentation

• Architecture — Code vs Data separation, portable design
• MCP-Server — 3 tools, hybrid search pipeline, cross-encoder reranking
• Technical-Deep-Dive — Thompson Sampling, vector database, cleanup pipeline
• Operations — CLI commands, health checks, auto-vectorization
• Evidence — Usage logs, decision funnels, proof the system works
• Comparisons — vs Claude Memory, CLAUDE.md, Generic RAG, Claude Code Skills

Full Interactive Documentation

View the complete interactive documentation with diagrams at: smartassist-ai.netlify.app