About This Notebook

This website exists to preserve engineering work over time.

It is not a resume and it is not a list of repositories. It records why a system was built, which constraints shaped it, what was measured, and what changed after the result was understood.

Questions Before Projects

Every project starts with an engineering question.

Can action recognition run without a backend? Can model loading overlap inference? Can a classical method outperform a deeper model under the actual constraints?

The question gives the work a durable boundary. Repository names, frameworks, and hardware will age. The reasoning remains useful.

Failures Belong Here

A failed experiment can eliminate an attractive but incorrect design.

The HotStart experiment made model loading slower. Recording the result, the baseline, and the causes is more useful than hiding it. Negative results prevent the same mistake from being repeated without new evidence.

Projects Over Technology Lists

Technology names describe materials, not engineering.

A project shows how those materials interact with memory limits, latency targets, data quality, deployment rules, and maintenance costs. That is why this notebook organizes work by questions and outcomes rather than by skills.

About Samuel

Samuel Diop is a Machine Learning Engineer with a PhD in Computer Vision.

His work focuses on GPU optimization, efficient inference, and machine learning systems, combining scientific experimentation with practical engineering.

He has experience developing computer vision systems, implementing model runtimes, and optimizing computational workloads under real-world constraints.

This website documents his technical work, research, experiments, and the engineering decisions behind them.

Explore the work · Read the notes · Research and publications