Production ML and anomaly detection
Detection and diagnosis for large fleets, telemetry and logs. Systems that point at a cause, not only an alert, taken from research to a deployed service.
How I work on thisMachine learning that survives contact with production.
I am Dr Yusufu Shehu. I design and build machine learning systems that have to hold up in production, and I give independent technical opinion on systems that have not. Most of my work has been in anomaly detection, forecasting and applied natural language processing for teams running large infrastructure.
Doctorate in particle physics University of Sussex, on the ATLAS experiment at CERN. MSci Physics, UCL.
Four first-author IEEE papers Peer-reviewed work on log message classification and anomaly detection.
Production at national-laboratory scale An anomaly-detection engine built for a supercomputing fleet of roughly 9,000 GPUs.
What I do
Most of the value in applied machine learning is lost in the distance between a result and a running system. I work in that gap. The methods are conventional on purpose. The care goes into the data, the evaluation, and the handover.
Forecasting and capacity planning
Headroom modelling and what-if forecasting for infrastructure and product teams who have to commit to a number.
How I work on thisApplied NLP and AI engineering
Classification and extraction for operational text: logs, alerts, tickets. Evaluated on your data, not a public benchmark.
How I work on thisIf your problem does not fit one of these, it may still fit the way I work. Describe it and I will tell you honestly whether it does.
Expert witness and litigation support
Independent, evidence-based opinion on how a model was built, what it was evaluated against, whether a technical claim stands up, and what the records show. The matters this covers include copyright and training data, model performance and failure, and professional negligence where a system has gone wrong. I give the same answer whoever is asking.
How engagements work
- Fixed-scope delivery. A defined piece of work with an agreed outcome and an agreed price. Best when you know what you need built and want it finished.
- Call-off advisory days. A block of days drawn down as you need them, for design review, method selection, or a standing second opinion on work your team is doing.
- Technical due diligence. An independent read on a machine learning system or a technical claim, for investors, acquirers, or a board that has been told something works.
Rates are quoted per engagement rather than published, because the shape of the work changes the figure. Ask and I will give you one within two working days.
Start with the problem, not the technology
Tell me what the problem or the matter is, what you have already done about it, and what a good outcome looks like. If I am not the right person, I will say so and point you somewhere better.
[email protected]