House of Agents
A digital House of Lords where you pick a real UK bill and twelve specialist AI-agent peers convene to debate it with retrieval-backed evidence, question witnesses, vote, and produce a scrutiny report or a draft letter to your MP.
Won $1,980 each in ElevenLabs credits at MongoDB.local London Hackathon (60+ teams).
MongoDB Atlas · Vector Search · Voyage AI · Google Gemini · ElevenLabs
What it does
You pick a real UK bill. Twelve specialist AI-agent peers convene, speak to each other, and debate with citations and retrieval-backed evidence. The Bishop can call you to the Witness Box. Then they vote. After the division, you get a Scrutiny Report, or it drafts a letter to your MP.
We wanted to see whether AI agents could simulate genuine multi-perspective debate on real legislation, grounded in actual evidence rather than hallucination.
How it works
Bills and supporting documents are stored and indexed in MongoDB Atlas. Before each peer speaks, they run a vector search query to pull relevant evidence from the index, so their arguments are grounded in what the documents actually say. Voyage AI handles the embeddings. Google Gemini drives the reasoning and language generation for each agent. ElevenLabs gives every peer agent a distinct voice, which makes the debate feel genuinely alive.
Each of the twelve agents has a distinct political background, area of expertise, and personality. They are not just taking turns to speak; they respond to each other, challenge positions, and cite evidence. The Witness Box mechanic lets the Bishop call the user in directly, so you can be questioned by the Lords mid-debate.
Result
Out of 60+ teams at MongoDB.local London Hackathon, we won the Best Use of ElevenLabs track, each taking home $1,980 in ElevenLabs credits.