MatLogica builds AADC, a just-in-time compiler for automatic adjoint differentiation in C++ and Python. AADC records a model as it runs, then compiles that execution into optimised machine code. The resulting kernel runs 6 to 1,000 times faster than the original code and computes the price and all first-order sensitivities with an adjoint factor below one. The price and its sensitivities together take less time to compute than the original code takes to calculate the price alone. AADC supports execution across Intel AVX2 and AVX-512, Apple Silicon and ARM architectures. Major Wall Street and global banks run AADC on more than $4 trillion of derivatives.

Founded by practitioners in quantitative finance and performance engineering, MatLogica applies the same technology beyond pricing and risk. Applications include training gas storage and hydropower dispatch policies using exact gradients, valuing and hedging variable annuities, and differentiating robotic trajectory optimisers.

The latest release adds AADC Tracer, which turns a model execution into a record that a validator can inspect, linking computational steps and hard-coded constants to their source file and line.

MatLogica founder and CTO Dmitri Goloubentsev presents on the Theta Stage at 12.25 on Wednesday 18 November. Visit our stand to learn more, or try the free community edition at matlogica.com.