RAG Cognitive Stack — Repository

This repository contains a production-ready scaffold implementing the five-persona RAG memory system (Archivist, Formalist, Synthesist, Strategist, Curator) plus routing, ontology tagging, benchmarks, UI, Dockerization, tests, and a LaTeX paper.

Contents

• memory_routing_schema.json — JSON Schema for memory routing records.
• prompts.py — Persona prompt templates.
• curator.py — routing implementation and curator logic (simulated LLM).
• ontology.py — vector tagging ontology.
• tagger.py — stub for embedding tagging projection.
• route_tests.py — pytest suite for routing.
• benchmark.py — synthetic failure-mode benchmark.
• app.py — minimal Streamlit UI for adding memories and seeing routing.
• samples/ — example .memory files.
• Dockerfile and docker-compose.yml — containerization.
• main.tex — LaTeX paper draft.
• requirements.txt — Python dependencies.

Usage (development)

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
pytest
streamlit run app.py

Notes

• LLM calls are stubbed/simulated. Replace the TODO markers in curator.py with real LLM calls (OpenAI/Anthropic/local).
• Embedding projection functions in tagger.py are placeholders; integrate your vector model (sentence-transformers or cloud embeddings) and train regressors to map embeddings to ontology scores.