Trust Me Bro

Inspiration

Prediction markets are having a moment. Kalshi and Polymarket now price everything from Fed decisions to Nobel Prizes, and the number on the screen gets quoted like a fact: "markets say 72%." But a price is just the last trade. It doesn't tell you whether someone with early information moved it an hour before the news, whether the same contract is trading 9¢ cheaper on the other venue, or whether a thin market got pushed around by one wallet.

In October 2025, the Polymarket market on María Corina Machado winning the Nobel Peace Prize spiked hours before the announcement. Everyone could see the price and almost nobody could see the receipts. Prediction markets basically say "trust me bro." We wanted to build the tool that checks and acts as a learning tool against this new type of investments.

What it does

• Trust Me Bro is a Bloomberg-style terminal for prediction markets, read-only and live across 400K+ Kalshi and Polymarket markets:
• Scanner. Every open market gets an anomaly score built from volume spikes, abnormal price moves, fresh-wallet concentration and timing against the news.
• Cross-venue arbitrage. We match the same event across Kalshi and Polymarket and compute the executable edge after spreads and fees, not just the headline gap.
• Agentic trading floor. Three AI analysts (The Quant, The Contrarian and The Risk Manager) each take a position on a contract, debate each other, and reach a consensus with a risk-capped stake suggestion.
• Crowd-reaction simulator. Drop a headline on a market and watch a swarm of AI trader personas react through a market maker, overlaid on what the real price actually did.
• Learn. Fifteen modules with live examples pulled from current markets, fee and stake calculators.

How we built it

Frontend: Next.js, Tailwind, shadcn/ui, Lightweight Charts and Canvas 2D for the swarm. Backend: FastAPI + Pydantic, with engines for anomaly scoring, mispricing, contract matching, wallets, swarm and consensus. Data: a poller pulls the full open universe from both venues (Polymarket's Gamma keyset feed and Kalshi's cursor pages), normalizes it into one schema, and recomputes signals off the event loop. Storage: Tiger Data (TimescaleDB). Every market and every price tick lands in a hypertable with hourly chunks.

Challenges we ran into

• Dead-flat markets produced fake 4-sigma events. A market that hasn't moved in a week has $\sigma \approx 0$, so a 1¢ tick looked like a once-in-a-lifetime event. We added variance floors (1¢ for moves, half the mean for volume) and a minimum of 24 hourly bars before scoring.
  • Live sports broke everything. In-game volume and venue lag look exactly like anomalies and arbitrage. We pull real game state from the venues and suppress flags from 3 hours before kickoff until the game ends.

## Accomplishments that we're proud of

• A live terminal over 400K+ markets that stays responsive.
• An anomaly score that flagged the Machado Nobel spike on replay, with a receipt a non-expert can read.
• A swarm simulator that spawns 5000 independent ai agents with different personas so you can check against what actually happened.

## What we learned

• Most apparent signal is noise. The hardest part wasn't detecting anomalies; it was not flagging things. Every threshold in our decisions log exists because something looked suspicious and wasn't.
• Time-series data wants a time-series database. Hypertables, compression and continuous aggregates turned "store every tick for 400K markets" from a scaling problem into a few SQL statements.

## What's next for Trust Me Bro

• Wallet-level clustering across markets to spot coordinated accounts.
• Connect and Directly make the trades for the user.

Built With

• fastapi
• nextjs
• tailwind
• tigerdata