πάντες ἄνθρωποι τοῦ εἰδέναι ὀρέγονται φύσει.
By nature, all human beings desire to know.
I am a machine learning engineer based in Paris, trained in applied mathematics. I hold a PhD in the field, and that background still shapes how I approach engineering problems: I tend to ask what a method actually guarantees, what assumptions it relies on, and where those guarantees stop matching the behaviour you observe.
My professional work spans statistical learning, numerical methods, retrieval, and production ML systems. Independently, I maintain open-source numerical software and collaborate on retrieval and LLM systems. This site is a place for that independent work: projects, technical notes, and questions I find worth exploring.
The writings here are mostly about statistics, signal processing, state space models, and machine learning. A recurring theme is that a method can be mathematically correct and still easy to read too strongly. A loss may measure one geometry while the application needs another. A guarantee may hold marginally while local behaviour changes. An estimator may be biased for a good reason. A finite optimization problem may be solved exactly while the population question remains open.
Most posts use small diagnostics, examples, and geometric viewpoints to make those gaps visible.
Projects · Writings · Email · GitHub · LinkedIn
Everything published here is personal and independent work, created outside my professional responsibilities using personal resources and public sources. It does not use or disclose employer or client work product, confidential information, internal systems, data, methods, or results.
Any discussion of industrial or applied settings is based only on general technical knowledge and publicly available references.