I coded my first machine learning model somewhere around 2005 in engineering school (a simple KMeans, in C) and that's when I fell in love with machine learning. Since then, - I did a PhD in computer vision (2010, using SVMs and SIFT-like features), - worked for a computer vision startup, - did some applied research at Criteo, mostly around recommender systems, using advanced logistic regression, GBDT, some neural networks, and AB-Testing, with large catalogs and billions of data points, - worked hands-on production systems at Datadog, building "simple" statistical models to detect anomalies in time series with low latency and high throughput - went back to a startup, managing a small team of machine learning engineers, building models and metrics for geolocation data. Now on a break.

I've started building agents but didn't go beyond tinkering. Never built an MCP server, a RAG, etc (though I did work on search engines in the past).

Tinkering with the LinkedIn portability API, which is badly documented. Authentication can be a nightmare.