How a Trading Firm Ships New AI Models the Same Day - Noyan Tokgozoglu, Optiver

Scaling AI in a high-stakes trading environment like Optiver requires balancing rapid innovation with strict financial accountability. Because trading decisions impact P&L within hours, the firm maintains an AI-native infrastructure that prioritizes modularity, on-premise security, and real-time performance. Key strategies include utilizing a hybrid model approach—leveraging both frontier and open-source models—and implementing an inference gateway to optimize costs and ensure resilience. To maintain efficiency, the team enforces "sane defaults" for token usage and employs a meta-harness layer that allows trading teams to configure their own experiences while maintaining portability. By fostering a self-service culture supported by forward-deployed AI engineers, the firm successfully integrates new AI paradigms into production flows, ensuring that every technological investment directly contributes to competitive advantage and measurable financial returns.