Machine Learning Engineer, AI Researcher, Founder / London
Temitayo Ayuba Abiona
I build production AI systems for speech, language, and enterprise decisions, then spend most of my time trying to prove them wrong.
Founder of Biona HQ, a speech intelligence practice, and Nora Health, which turns a clinical consultation into a structured note without the audio ever leaving the room. My work sits where research has to survive contact with production.
Selected outcomes
- Less manual effort per model retrain after the platform rebuild
- 90%Less manual effort per model retrain after the platform rebuild
- Annual infrastructure spend removed with no loss of delivery pace
- £100kAnnual infrastructure spend removed with no loss of delivery pace
- Grounding quality on a production generation pipeline
- 87%Grounding quality on a production generation pipeline
- Patient audio recordings leaving the clinician device
- 0Patient audio recordings leaving the clinician device
Selected work
Systems built to be checked, not admired
Projects where the engineering problem was not building the model. It was establishing what the system was allowed to claim.
Research
An active programme, not an archive
Current work on verifiable generation, speech models at the edge, and why benchmark performance predicts production performance so poorly.
Lab
Work in progress, including the parts that failed
Benchmarks, prototypes, and experiments published while they are still open questions.
Insights
Writing on what the work actually taught me
Notes on evaluation, speech systems, and building with language models, written for engineers.
About
Statistics first, then speech, then everything that breaks in production
I started in statistics and anomaly detection, where being wrong is expensive and the rare event is the entire problem. That framing carried into speech, language models, and the infrastructure underneath them.
Contact
If you are working on something where being wrong matters, I would like to hear about it.
I am open to consulting engagements, research collaborations, and conversations that do not have a clear outcome yet.