AI technical review and advisory

Independent technical review of AI systems, from a working researcher.

I am Dr Riad Ibadulla, a Lecturer in Computer Science at City St George’s, University of London, with a PhD in deep learning architectures for optical AI accelerators and industry experience as a Deep Learning Engineer at Optalysys. I review and advise on the technical work of smaller AI companies and regulated organisations: how models are built, how they fail under pressure, and how they handle data.

Riad Ibadulla in a dark suit and blue tie, arms folded, standing beside a white column.

Specialisms

Efficient deep learning architectures

Designing and adapting neural networks to run within tight limits on compute, memory and latency. My doctoral research produced FatNet, Fat-U-Net and ConvShareViT, architectures for free-space optical AI accelerators, the latest published in IEEE Transactions on Neural Networks and Learning Systems.

Publications

Adversarial robustness with formal methods

Testing, and where possible proving, how models behave when their inputs are deliberately manipulated. PROTECTION, presented at a SAFECOMP 2025 workshop, uses SMT-based verification to give formal robustness guarantees for a machine-learning intrusion detection system.

Read about PROTECTION

Latest Technical Notes

All notes

Published technical reviews of companies’ public AI work, scored against seven criteria.

No notes have been published yet.

From hackathon to government tool

In July 2023 my team won the Generative AI Hackathon run by the 10 Downing Street Data Science team through Evidence House, with a prototype that used a large language model to modernise the ministerial red box. We presented it to Minister Alex Burghart at 10 Downing Street, and I was invited to the meeting that set up the follow-on project. The concept was then developed by the government’s Incubator for AI into Redbox, which was used by more than 6,000 civil servants.

Sources: City St George’s news, Incubator for AI, Redbox reflections.

Discuss a review

The quickest way to reach me is by email at [email protected]. A short description of your system and what you would like looked at is enough to start.