Hi, I'm Leo.
I'm an AI engineer specialising in real-time voice AI, training-data engineering, model fine-tuning and evaluation. I build working systems: things people can talk to, and data that measurably makes models better.
I'm the founder of Mentara Persona, an AI voice persona product. I built the whole stack myself: the live speech pipeline, the persona and behaviour layer, accounts and billing, the server side and the desktop app. Its flagship persona, Rebecca, has been on enough live calls to fill 100+ YouTube Shorts.
Running a live product taught me what matters in practice. Latency, interruptions and background noise decide whether a voice AI feels alive or broken. Most model problems turn out to be data problems. And automated pipelines need a human checking their output.
Alongside client work I'm developing my own AI technology, starting with an AI-powered data-preparation system.
Principles
Demos over decks
I'd rather show you a system working than describe what it could do.
Defined deliverables
Every engagement states what you get, how it's evaluated and when it's due.
Honest limits
Evaluation reports cover regressions and limitations, not just wins.