About
Models that
reach production.
I'm a machine learning engineer with an MEng in Data Science, based in Tunisia. I build and ship full-stack ML systems solo — computer vision pipelines, RAG and agent systems, and the backends behind them. A model that never ships isn't finished, so I own the whole path from research to product.
Timeline
The Road Here
MEng, Data Science
Graduate Engineer
Master of Engineering in Data Science. Built the foundations I use daily — statistics, deep learning, and the maths under modern ML — with a bias toward things that actually run, not just derivations on paper.
Into Production ML
Machine Learning Engineer
Moved from notebooks to shipped systems. Learned that a model is maybe 20% of the work — the rest is data pipelines, serving, evaluation, and the backend that makes it usable by real people.
Computer Vision & RAG
ML Engineer
Went deep on two fronts: chained computer-vision pipelines (detection, segmentation, image registration) and retrieval-augmented generation over messy, unstructured data. SafeScan AI and the research knowledge base came out of this stretch.
Lead ML Engineer
Shipping ML Products Solo
Owning ML systems end to end — models, pipelines, backends, and the product around them. Also running a Bittensor subnet for decentralized lesion detection. The throughline: research that reaches production, shipped by one person.
Capabilities
Skills & Stack
ML & Computer Vision
LLMs, RAG & Agents
Backend & MLOps
Full-Stack & Product
Direction
What I Focus On
Computer Vision Pipelines
Chained pipelines where detection, segmentation, and registration feed each other — built to hold up on external data, not just a clean training split.
RAG & Agent Systems
Retrieval over unstructured, real-world data. Grounding answers in ingested sources instead of model priors — clean extraction and chunking first, model second.
Production Backends
The unglamorous 80%: serving, data pipelines, and APIs that turn a model into something people can actually use. Python and FastAPI, built to ship.
Shipping Solo, End-to-End
From model to product, owned by one person. No handoff between "the ML" and "the app" — the whole system is the deliverable.