Large language models and alternative data often dominate conversations about deep learning in trading. At high-frequency horizons, Karun Rao looks instead at the order book, where models learn from noisy, event-based market data whose structure is hard to capture by hand.

A conversation with NVIDIA on how accelerated computing is changing research and engineering in trading.

Optiver’s software engineering function was profiled by The Pragmatic Engineer, a leading technology publication on Substack.

Latency and research iteration pull in different directions. Software engineer Daniel B. explores how three generations of trading systems evolved to balance both.

For the past several years, the AI conversation has largely centered on models: which are the most capable, which will dominate and how quickly they will improve.

Instead of throwing agents into an SDLC that wasn’t built with them in mind, we’re trying to do something different: design the development lifecycle around them from the ground up. Below is an example of how we shrank a block of work we take on when we connect to a new exchange. Over the last few months, we've been running an experiment. If you take agentic coding seriously, what does it look like?