What if you could call 1,000 people and ask them what they're watching tonight, then bet on that data before anyone else knows about it?
That's basically what this does.
Prediction markets are awesome, but everyone's working with the same public information. Polymarket, Kalshi, whatever - they're all reading the same polls, the same news, the same Twitter sentiment.
We figured out how to generate actual unique alpha by mass-calling real people with AI agents and extracting insights you literally can't get anywhere else.
Want to know which TV show is gonna trend next week? Call 500 people and ask what they're planning to watch. Want to predict box office numbers? Ask people if they bought tickets. Political sentiment in swing states? Just... ask swing state voters.
Then take that data and trade on it before the market catches up.
The pipeline is surprisingly elegant:
-
Conversational AI Cold Calling - We spin up ElevenLabs conversational agents that sound legitimately human. These agents make outbound calls through Twilio to real phone numbers.
-
Natural Language Data Extraction - The AI agents have dynamic prompts (powered by Groq/OpenAI) that let them have actual conversations. People don't realize they're talking to an AI half the time. The agents extract structured data from natural conversations.
-
Sentiment Analysis at Scale - All call transcripts get fed through Perplexity AI for deep sentiment analysis and pattern recognition. We're aggregating hundreds of data points into actionable insights.
-
Prediction Market Integration - Direct Kalshi API integration lets us analyze markets, identify mispricings, and see where our unique data gives us an edge.
-
Real-Time Dashboard - Next.js frontend that shows live market data, call results, sentiment breakdowns, and lets you launch new calling campaigns on the fly.
- ElevenLabs Conversational AI - Ultra-realistic voice agents that can handle dynamic conversations
- Twilio - Phone infrastructure for making thousands of outbound calls
- Groq SDK - Lightning-fast LLM inference for real-time prompt processing
- OpenAI API - Fallback LLM for more complex reasoning tasks
- Perplexity AI - Sentiment analysis and online research synthesis
- Kalshi API - Direct prediction market data and (eventually) trading
- Next.js 14 - React frontend with server components
- Fastify - High-performance Node.js backend for handling WebSocket streams
- WebSocket streams - Real-time bidirectional audio streaming between Twilio and ElevenLabs
Traditional prediction markets rely on:
- Public polls (everyone has these)
- Social media sentiment (easy to manipulate)
- News articles (lagging indicators)
We're generating:
- Primary source data from actual human conversations
- Pre-trend insights before behavior shows up in public data
- Granular demographic targeting (call specific zip codes, age groups, etc.)
- Sentiment depth that surveys can't capture
Some ideas we've tested or are testing:
- Entertainment markets - "What show are you binge-watching this weekend?"
- Consumer behavior - "Have you bought any tech products recently?"
- Political sentiment - "How do you feel about [candidate/policy]?"
- Local events - "Are you planning to attend [event]?"
- Product launches - "Have you heard of [new product]?"
Anything where consumer intent predicts market outcomes.
Look, we're aware this is kind of dystopian. Using AI to call people en masse for financial gain probably isn't what the future was supposed to look like.
But prediction markets are fundamentally about information aggregation, and this is just... really good information aggregation. Plus people can always hang up.
Built because we realized prediction markets are just arbitraging information asymmetry, and phone calls are asymmetric information at scale.