Skip to the content
ProphETF

How it works

You type an idea. An AI agent picks three stocks for it, live prediction-market odds set their weights, five more agents vote, and every call is scored a week later. Nothing is bought or sold: these are paper portfolios.

From idea to basket

  1. 1

    Your idea

    A sentence, in any language.

  2. 2

    Gemini agent

    Picks 3 stocks from 40, or asks you to clarify.

  3. 3

    Polymarket odds

    Live odds of the events those stocks depend on.

  4. 4

    Weights

    Equal weights, tilted by this week's odds moves.

  5. 5

    5 AI votes

    Five agents vote on each stock and on the basket.

  6. 6

    Track record

    Every call is scored against the S&P 500 a week later.

  7. 7

    Lab

    Paper-trade the basket and test event scenarios.

What it is built with

In the product

  • Google Gemini

    One agent picks the three stocks, five agents vote on them, and a manager's note is written when a basket is saved. An answer is reused until the odds behind it move 3 points.

    Models
    —
    Gemini calls today
    —
    Answers reused today
    —

  • Tiger Data

    Postgres with TimescaleDB keeps every odds reading, the stock closes and each panel decision, so charts and the track record are plain SQL.

    Odds readings (odds_ticks)
    —
    Stock closes (stock_closes)
    —
    Panel decisions
    —
    Latest odds reading
    —

  • Polymarket and Kalshi data

    Public prediction-market odds, read-only. No keys, no accounts and no orders: the odds show what is at stake, not which way to bet.

    Markets tracked
    —
    Access
    Read-only, no keys

  • Stock prices

    Daily closing prices from Yahoo Finance, loaded with Hao's Python code. They score every panel decision against the S&P 500.

    Source
    Yahoo Finance
    Daily closes stored
    —

How we built and run it

  • Claude Code

    Wrote and tested the backend, the stock agent, the voting panel and this website, working from Lucas's specifications.

  • Vercel

    Hosts the website and the API as one Next.js app, deployed from the command line.

  • cron-job.org

    Calls our odds job every minute, which stores one reading per tracked market in Tiger Data. The time below is the newest reading it wrote.

    Job
    /api/jobs/tick, every minute
    Last tick
    —

  • GitHub and GitHub Actions

    The repository, and the tests of every part run on each pull request.

  • Next.js and TypeScript

    The website and the API routes. The stock agent itself is plain JavaScript, shared with a local server.

  • Python

    Hao's research (the event study and the backtests) and the weighting method. The live weights are a TypeScript port, kept identical by a parity test.

Loading the live numbers…