Inspiration
Traditional focus groups are slow, biased, and expensive. We wanted to build a "Digital Twin" of the internet that doesn't just search for keywords, but understands the underlying sentiment and tribal dynamics. By observing how Reddit, YouTube, and X users discuss existing tech, we can predict how they’ll react to a product that hasn't even launched yet.
The goal is to bridge the gap between messy internet discourse and structured market research by simulating how real humans—grouped into "Tribes"—react to new product ideas.
How It Works The system follows a high-fidelity data pipeline (from ECE-style signal processing to AI reasoning):
Ingestion: We scrape real-time data from subreddits and tech communities.
Normalization & Chunking: Raw HTML is converted into clean Markdown and split into "AI-sized" pieces while preserving metadata.
The Knowledge Graph: Using Gemini, we extract Entities and Relationships.
Community Detection: We apply the Leiden Algorithm to the graph to find clusters of people with similar values.
Persona Simulation: We feed these communities into LLMs to generate Synthetic Personas. You can then "chat" with these personas to see if your new product hypothesis holds water.
Why It’s Unique: Unlike a simple chatbot, this uses RAG (Retrieval-Augmented Generation) and GraphRAG. This means when a persona speaks, they aren't just hallucinating; they are citing actual evidence from the Knowledge Graph you built.
Log in or sign up for Devpost to join the conversation.