Quants and developers who have done the integration before. We benchmark, run a proof of concept, sit with your team during integration, and teach AAD and HPC when that is what is needed.
An introduction and training if your team is new to AAD, a proof of concept on your own model, support during integration, and help in production afterwards. Pick what you need; most clients start with the proof of concept.
If you are not sure AADC pays off on your problem, whether that is derivatives pricing, XVA or risk, or you want it measured against another AAD approach, we run the benchmark on your use case and give you the numbers.
A few weeks on one of your own models. Free, because it is the fastest way for both of us to find out whether AADC is worth your time on that pricer or that risk run.
We sit with your quants and developers while AADC goes into the pricing, risk or XVA system, so the questions get answered the day they come up.
A lecture on high-performance computing or AAD, pitched for the audience you have, whether that is a quant team that wants the mathematics or a management group that wants the economics.
In person or remote, on AAD and HPC practice for pricing, Greeks and XVA. Hands-on, on code that looks like yours.
A call to understand what you are trying to speed up or differentiate
Measure it on your own model
Your team learns the patterns before the integration starts
Side-by-side support during deployment
Help in production, and further tuning when the workload changes
Ten problems that arrive in our inbox, in the words they arrive in, and what we do about each.
“Barrier and autocall Greeks are noisy if we bump and biased if we smooth, and the smoothing width is tuned per product.”
Smooth once, consistently, inside the library, and take exact Greeks from one pass. Validated for autocallables in production.
Greeks for autocallables →“The model calibrates on every call, so the adjoint stops at the solver.”
Automatic IFT differentiates through the calibration with no change to the solver.
Automatic IFT →“Risk runs overnight; the desk wants it at eleven; nobody will rewrite the library.”
Record the library once and replay on demand: a million FX trades in 0.4 s on unmodified ORE.
ORE live risk →“The tape does not fit in memory on long paths, American Monte Carlo, or a thirty-year book.”
The recording is compiled once; each scenario then needs a fixed workspace, not its own tape.
AAD for American Monte Carlo →“Hand-written adjoints cover part of the solver and break with every model change.”
A typedef makes the whole solver differentiable; the adjoint is generated from what ran.
How it works →“Validation reconstructs what the code does by hand, and the documentation drifted at the last rebuild.”
AADC_TRACE=1: every step tied to its line, every constant and branch listed, the report drafted from the record.
AADC Tracer →“The research is in Python; production is assumed to need C++.”
Compile the Python model. Python stays the source; the kernel runs at C++ speed.
Python Accelerator →“The library is tied to a vendor's roadmap, or it is ours and costs a team to maintain.”
AADC runs on QuantLib, ORE or your own library. The code stays yours.
Open-source libraries →“Reinforcement learning took four hours on our dispatch problem and still found a worse schedule.”
Train the policy through the recorded simulation with exact gradients: 39 s to $2.19M, where RL reached $1.07M.
Gas storage dispatch →“Our trajectory optimiser's gradients get slower with every variable we add.”
Reverse mode through the same simulation code: 30× faster than Drake's forward-mode AD at 1,004 variables, identical accuracy.
Differentiable simulation →Contact us to discuss your requirements and learn how our professional services can accelerate your AAD implementation.
Benchmarking on your workload, a free proof of concept on one of your models, support during integration, guest lectures on HPC and AAD, and training workshops in person or remote. The people doing it are quants and developers who have done the integration before.
Yes. A proof of concept runs a few weeks on one of your own use cases, such as a pricer, an XVA calculation or a risk run, so you see the speed-up on your code before committing to anything.
MatLogica integration consulting includes side-by-side support working directly with your quantitative developers and quants throughout the AADC integration process. Our experts provide technical guidance, best practices, troubleshooting, and hands-on assistance to ensure successful deployment into your existing derivatives pricing, risk management, or XVA calculation systems.
MatLogica training workshops cover AAD theory and applications, HPC optimisation techniques, AADC integration best practices, performance tuning for quantitative finance, code generation concepts, derivatives pricing with AAD, Greeks calculation methods, and XVA computation optimisation. Workshops are available in-person or remote, tailored for quantitative developers and analysts.
MatLogica guest lectures on HPC and AAD are suitable for quantitative analysts, developers, risk management teams, and management at financial institutions. Lectures are tailored to organizational needs and can cover technical details for developers, strategic benefits for management, or practical applications for quants. Both in-person and remote formats available.
MatLogica benchmarking service evaluates AADC performance for your specific domain area (derivatives pricing, XVA, Monte Carlo, etc.) and provides comparison with alternative AAD approaches. The benchmarking process uses your actual use cases and data to provide data-driven performance metrics, helping you make informed decisions about AADC adoption.