Posts

I Built a Search Agent. Then I Spent Months Teaching It When Not to Answer

Image
  This article is the write-up of my Talk at the Paris Machine Learning meetup The hardest feature in production RAG is not retrieval. It is judgment. A funny thing happens when you build your first search agent. At first, it feels magical. You take a messy pile of knowledge: PDFs, internal templates, business rules, operational instructions, store information, support documents. You chunk everything. You embed it. You retrieve the closest passages. You send them to an LLM. The model writes an answer. In a demo, everyone smiles. Then real users arrive. They ask half-questions. They use old names for new processes. They ask about rules that changed last week. They mix several requests in one sentence. They paste personal information. They ask things that look answerable, but are not actually supported by the documents. And the system does exactly what you built it to do. It answers. That was the uncomfortable lesson behind my Paris Machine Learning Meetup talk: A good search agent i...
Read More >>

The Complete (& simple) Guide to Agentic AI in 2026: From Chatbots to AI Coworkers

Image
The Complete Guide to Agentic AI in 2026: From Chatbots to AI Coworkers A plain-English introduction to what AI agents are, how they work, and why everyone is suddenly talking about them. A few years ago, the main question was simple: “What can ChatGPT answer?” Then the question became: “What can AI help me write, summarize, code, or analyze?” Now, in 2026, the question is shifting again: What can AI actually do for me? That shift is why everyone is talking about agentic AI . The term sounds technical, maybe even a little overhyped. But the basic idea is simple: A chatbot answers…. An AI agent acts. A chatbot waits for your prompt. An AI agent can take a goal, break it into steps, use tools, check what happened, and continue working toward the result. That does not mean AI agents are magic. It does not mean they are fully autonomous employees. And it definitely does not mean you should let them do anything without supervision. But it does mean we are moving from AI that only talks to A...
Read More >>

AI: "It's raining cats, dogs and pitchforks"

Image
A survival guide for the AI flood nobody asked for but everyone is drowning in Let me start with a small historical detour. Bear with me. When Gutenberg finished tinkering with his printing press around 1440, the world did not just get books. It got flooded with pamphlets, religious texts, political manifestos, bad poetry, and the 15th-century equivalent of LinkedIn thought leadership posts. Scholars at the time complained that there was simply too much to read. Too much information. Too many opinions. Sound familiar? Image: Reproduction of a Gutenberg-era press at the Printing History Museum in Lyon, France. Photograph by George H. Williams, via  Wikimedia Commons . Fast forward to the late 1800s. Electricity arrives. Not just as a concept, but as a roaring industrial wave. Factories rewire. City streets get lit up. People panic about whether sleeping near electrical wires would kill them in their beds. Newspapers ran think pieces about "the overstimulation of modern life." ...
Read More >>

7 Charts That Lied to Your Face: Why Bad Data Visualization Is the Class My Students Never Miss

Image
For the past few years, I have been teaching data visualization to data science students at Université Paris Cité at the IUT de Paris — Rives de Seine . And while teaching Python coding is a relatively simple thing, I have decided to shift the course toward a more useful and pedagogical angle: how to make data speak, and how to make good charts, the correct way. One of the chapters I spend around 25% of the course time on is bad data visualization: mistakes made with charts, misleading visual design, and how a graph can quietly manipulate the reader. This is the part of the course where I get 100% attendance and around 97% active interaction from my students. Yes, I am a statistician, and I keep track of a lot of metrics. It is probably because this is the single course section where they are not required to open their computer, nor to work on a practical exercise. I start the chapter by dimming the light of the room. I launch my slides. No explanation. No theory. No code. Just charts...
Read More >>