Availability: Open to opportunitiesNouakchott, Mauritania
Bechir Mady
I build end-to-end data systems — from real-time ingestion through to a model served behind an API — and I measure what I ship rather than assert it.
MSc student at SupNum, Nouakchott. Spark and Delta Lake lakehouse, forecasting with calibrated intervals, multimodal RAG with verifiable citations, geospatial machine learning on Sentinel imagery.
Introduction
Overview.
Data and machine learning engineer, comfortable from the source through to the served model: real-time ingestion on Spark and Delta Lake, models whose uncertainty is measured rather than assumed, retrieval with verifiable citations, and machine learning on satellite imagery. I deliver with tests, orchestration and a Docker image — not with an isolated notebook.
Data Engineer
Machine Learning Engineer
MLOps & Platform
Geospatial & Vision
Work / 9 projects across 4 categories
Projects.
Each category groups the projects that share a way of working. Open one to see its projects, what they measure, and their code.
Path / 4 milestones
What brought me here.
Master's thesis — rice yield estimation
InsightEx × SupNum × ISET Rosso
- An end-to-end study of the 2025 hot off-season campaign, from the Sentinel-2 pixel through to the observed yield.
- An exact join of all 102 Trarza plots to their harvest yield, across two datasets that keyed them differently.
- A rigorous comparison of several architectures, written up as a nine-chapter thesis that runs as a notebook.
Research paper submitted
I2COMSAPP'26 — Combinatorial Optimisation
- One-to-One SkewGRAM, in a team of four, supervised by Dr. Mohamed Lemine Ahmed Sidi and Dr. Hafedh Mohamed Babou.
- The Skip-Gram paradigm transposed to an NP-hard problem over two coupled graphs.
- Decoding guided by exact dynamic programming, then extraction by incremental Union-Find.
MSc, Data Engineering & AI
Institut Supérieur du Numérique (SupNum), Nouakchott
- Big data, machine learning, distributed architectures and natural language processing.
- Most of the projects in this portfolio started during this programme.
Self-directed platform projects
Independent work
- A real-time lakehouse, calibrated exchange-rate forecasting, multimodal retrieval with verifiable citations.
- Each one delivered with its tests, its orchestration and its Docker image rather than as an isolated notebook.
Open to opportunities
Let's talk.
Available for an internship, a data or ML engineering role, an open-source collaboration, or simply a good technical conversation. I answer every message — in Arabic, French or English.