PolyMind

Multi-agent RAG system exploring collective reasoning and bias in AI decision-making

A multi-agent Retrieval-Augmented Generation (RAG) system that explores collective reasoning and cognitive bias in AI decision-making. Five specialized cognitive agents — Optimist, Skeptic, Historian, Forecaster, and Judge — independently retrieve from a shared knowledge base and synthesize a consensus answer.

The system includes a full evaluation framework using RAGAS and DeepEval for hallucination detection and response quality assessment, as well as a REST API and Python SDK.

Tech Stack: Python, FastAPI, LangGraph, LanceDB, OpenAI, Anthropic, RAGAS, DeepEval, React, TypeScript

View on GitHub