Experiments.

The questions I’m exploring, with the measurements, write-ups, and simulations behind them.

Benchmark database

What changes local inference performance?

Compare measured prompt processing and generation across GPUs, quantisation, MTP, context depth, and concurrency.

Research write-up

Separate learning the rules from learning a strategy, then evaluate how training data changes the model’s play.

Simulation project

A repeatable limit-order-book simulation for changing trader behaviour while keeping the starting conditions fixed.

Looking for the tools behind the work?