Data, ML & Inference

I build ML systems that make decisions — recommendation, forecasting, conversational AI — and this site is where I write about how machines learn to decide, from statistics up to reinforcement learning.

My degree is in statistics, so “inference” has always meant two things to me: the conclusion a model draws, and the service that has to deliver it. I work on both ends — whether the estimate holds up, and whether the endpoint stays up. A model that is right in a notebook but unreliable in production hasn’t shipped; it’s just been demonstrated.

That standard runs through everything I’ve built: a recommendation and sentiment system serving live predictions inside a campus events app, a chatbot handling client intake and common legal questions, and time-series forecasts of sales and market trends for early-stage startups. At Ertaqy, I design the Python backends behind AI features — inference serving, API design, and the statistical evaluation that tells us whether any of it actually works.

Two convictions shape how I work. Machine learning is one component of a wider software system, not the centre of it — so I optimise for clarity, reproducibility, and system health before leaderboard scores or novelty. And the theory underneath deserves to be written down properly — the essays here climb from probability and optimization toward reinforcement learning, because deciding well is a harder problem than predicting well.