MIRRA - Bedrock-powered RAG system
Built a Bedrock-powered RAG system that simulates real consumer behavior for AWS Bedrock x Columbia Engineering Hackathon.
Project Overview
Co-created MIRRA, an AI-powered consumer-simulation platform designed for pre-market product validation, presented at the AWS x Columbia Engineering Hackathon. Our system replaces slow, costly focus groups by using Amazon Bedrock + RAG to simulate realistic consumer behavior across demographic personas. Designed a full pipeline that ingests product descriptions, normalizes attributes, retrieves relevant persona clusters, and generates tailored synthetic feedback using LLM-PAG reasoning. Built an interactive dashboard that visualizes purchase likelihood, persona distribution, demographic breakdowns, and sentiment insights—transforming weeks-long research cycles into days. Demonstrated business impact through ~60% lower cost, ~80% faster testing cycles, and ROI-positive market testing compared to traditional focus-group workflows.