The problem ZKML Options Liquidator solves

Options trading on DeFi platforms requires intricate calculations for determining liquidation thresholds, a task complicated further by the need for reverse-engineering "Implied Volatility" from market trades. Traditional methods rely heavily on the Black-Scholes formula, posing trust and transparency issues when executed off-chain. Our approach overcomes these challenges by integrating AI/Machine Learning with Zero Knowledge Proofs (ZKP) to ensure accuracy and verifiability of liquidation operations.

Challenges we ran into

Proving System Limitations: Initial attempts at step-by-step AI inference emulation proved prohibitively expensive. Adoption of the EZKL system, optimized for matrix multiplication in ML, presented a viable alternative.
Data Input Validation: The absence of ZK-friendly oracles necessitated a novel approach to validating input data. We achieved this by verifying blockchain transaction signatures and generating ECDSA proofs with Brevis.
Proof System Incompatibility: Integrating disparate proving systems posed a significant challenge, addressed by employing a common Poseidon hash to ensure consistency across proofs.