Embersity
Prediction markets need better questions, colder-start liquidity, and resolution people can trust. AI belongs in those loops, not as decoration.
Context
Prediction markets don't fail only at trading - they fail at question design, liquidity cold-starts, and resolution trust. Embersity attacks all three, with AI in the loop at creation and settlement.
What I did
Implemented core: binary market creation; a signed CLOB order lifecycle; on-chain collateral reservation so resting orders are always backed; runtime matching; on-chain settlement; isolated leveraged positions with liquidation; the shared EMBERVault with borrow reserve, bad-debt accounting, open-interest caps, and transparent FIFO payout queues; resolution primitives with evidence hashes and dispute windows; AI-assisted market-spec generation via structured outputs; Ponder indexing; bots; admin operations; and a full trading frontend. Designed direction: Embie as a conversational market-quality engine, risk-tiered leverage up to 100x for eligible markets, and AI consensus resolution through independent resolver agents, adversarial review, confidence scoring, challenge windows, and on-chain settlement reports.
How it works
Embie creates it, Martingale bootstraps it, traders leverage it, AI consensus resolves it - with the vault as the solvency spine: exposure caps, utilization limits, and payout queues that make settlement constraints transparent instead of hidden.
Impact
A production-shaped stack - not a contract demo - and a working thesis for prediction markets as AI-native event derivatives.