Services
Services
Three review packages for smaller AI companies and regulated organisations. Each one ends with a written report that explains what I found, how serious it is, and what to do about it.
Durations below are typical elapsed times. The actual time, and so the price, depends on the size of the system, how deep you want the review to go, and how much access you can give to code, models and documentation.
Architecture and efficiency review
For teams whose models need to be faster, smaller or cheaper to run, or who want an outside view before committing to an architecture.
Scope
- Model architecture and the reasoning behind the main design choices
- Training and evaluation pipeline, including how results are measured and reproduced
- Inference cost: latency, memory, throughput and hardware fit
- Options for reducing cost, such as architectural changes, pruning, quantisation or distillation
Adversarial robustness audit
For systems where a wrong answer is costly, or where someone may try to make the model fail on purpose.
Scope
- Threat model: who might attack the system, how, and what they would gain
- Testing against adversarial examples and input perturbations relevant to your data
- Where the model is small enough, formal verification of robustness properties
- Recommended defences and monitoring, ranked by effort and benefit
Privacy posture assessment
For teams that train on or process personal or sensitive data and need to know where the risks are.
Scope
- How data is collected, stored, used for training and retained
- Risks of models memorising or leaking training data
- Use of third-party model APIs and what data leaves your control
- Practical privacy-preserving options, such as anonymisation, local inference or federated approaches
ISO/IEC 42001 readiness
ISO/IEC 42001 is the international standard for AI management systems. I can carry out a gap analysis of your current practices against its requirements and help you prepare for a formal assessment.
This is preparation, not certification. Certification can only be issued by an accredited certification body.
How reviews are scored
Reports use the same seven criteria as the published Technical Notes, weighted to suit your product and setting:
- Architecture and model design
- Code quality and engineering practice
- Robustness and adversarial resilience
- Efficiency and resource use
- Data handling and privacy posture
- Documentation and transparency
- Deployment and production readiness
To discuss which review fits, email riadibadulla@gmail.com.