The journey

I didn't want the clean, flattering version people write two years later, where every detour looks planned. So this includes the doubts, the lucky breaks and the things that went wrong. It's not finished, and I keep adding to it as it goes.

A tennis scholarship brought me to the US. I was recruited to play NCAA Division II at Menlo College in California. The sport was literally the door the whole story walked through. Season record: 25 matches, 16W / 8L / 1 draw, 64% wins.

Match point. The team reaction is the part you remember.

Sport at this level doesn't teach you "business lessons." It just trains habits. Showing up when you're tired. Preparing for something you might lose. Recovering from a bad set without letting it ruin the next one. Those habits carried over. Not as a metaphor; as a daily practice.

Marathon personal best: 2:52. Multiple 100 km ultramarathons completed. The progression from short-distance tennis explosiveness to ultra-distance patience taught me more about myself than any competition result.

I run because it's the one part of my life where nothing can be negotiated. A 2:52 marathon or a 100 km trail doesn't care what else happened that week. What endurance actually gave me is quieter than any race result. Knowing how to keep going when the interesting part is long over. Knowing how to recover instead of just pushing. And knowing that most hard things stop being hard the moment you decide they're simply what you're doing today.

A year there costs around 110k USD; my agency negotiated a 70% scholarship. Without the racket, that door wouldn't have opened so early.

The timing wasn't an accident either. Everything I was working on pointed west. The deeptech founders I interviewed kept describing the US as the place where a technology is judged on what it could become. Not on the paperwork behind it. I wanted to see that from the inside rather than from a report. The plan was to come back after graduating and work on OPT. That part hasn't gone to plan yet, and I tell the whole story on the Failures page.

The first few weeks were disorienting. Everything moves faster. The unwritten rules are different. And you realize very quickly that what made you credible in France doesn't automatically translate.

One thing I'd tell anyone arriving: take your first week and spend it meeting as many people as you possibly can. Don't plan, don't optimize, just say yes to everything. It's the fastest way to understand the culture, absorb the energy, and figure out where you actually fit. The relationships you build in that first week often end up being the most important ones.

I also went to everything else I could get into, like the Startup World Cup run by Pegasus Tech Ventures. It was the same exercise every time: present the work, defend it, take the pushback. I did it in a second language, in front of audiences used to fast, direct answers. It taught me more than any amount of reading.

Volunteering is the cheapest ticket into rooms you have no other reason to be in. More than 5,000 miles from home, the French community turned out to be a piece of home. Alumni, founders who had made the same crossing, rooms where the accent, the references and the ambition all lined up. A stranger could become a contact in one conversation, and that familiarity was worth more than any formal introduction.

On a court in California, you just play, and afterwards you talk. The racket did more for my network in a year than any amount of cold outreach would have. It put me across the net from people I had no other reason to meet.

What made it actually work was one boring habit I forced on myself: I wrote everything down. Every decision, every spec, every next step, captured so nobody was ever stuck waiting on a call to move. We shipped the financing decks, the CRM, the pricing, on the dates we set, from three different time zones. I learned more about running a team from that constraint than from any class.

That shift was uncomfortable at first, and then it became the most useful thing I brought back.

A whole week of episodes might collapse into ten usable ideas. How to read an AI announcement like an analyst. Why vertical, workflow-integrated products beat general ones. It's a slower way to listen and a much faster way to learn.

It's published on SSRN, was presented in Oakland and in Nice, and was picked up by VC funds once online. My Menlo capstone on delayed AI-generated trading decisions is the second published piece. Two more are in progress. One on Europe's scale-up gap. One with Amine Lahiani, on whether accessible AI can help European DeepTech industrialize. The academic work and the advisory feed each other: the research asks the questions, the fieldwork tests them.

So I started describing what I wanted in plain language. AI drafted a first version. I tested it, broke things, fixed them, broke them again. Progressively I started understanding what was happening behind the code. Not deeply, not like a developer, but enough to shape the output rather than just accepting it.

The CRM became my testing ground, but it didn't stop there. I rebuilt this website myself. I wired up prospecting tools for the startup. I ran the regression work for the research myself. OLS models in Python with statsmodels. The full econometric pass in Stata for the VC and ESG paper, log transformations, board variables and all. I experimented with blockchain at Paris Blockchain Week, and with every new AI coding tool as it landed. The pattern was always the same: start building before I fully knew how, and let the thing teach me. The messy spreadsheets turned into one consolidated system: thousands of contacts, deduplicated, tagged, with a live priority score and a dashboard.

Then I added an AI layer on top: ranking who to reach out to, drafting first messages, remembering when to follow up. An afternoon of work for something that would have taken weeks, or an outsourced developer, a year earlier.

I didn't become a developer. I learned to build.

What changed wasn't that I could suddenly write software. It was the distance between "I need a tool that does X" and having that tool. It went from months and a budget to an afternoon. Ideas I'd previously have abandoned because building them was out of reach became things I could just try.

And when a company needs a tool to track something, I don't automatically assume it has to be bought.

The progression mattered more than any single certificate. First Bloomberg Market Concepts and Finance Fundamentals, to learn the terminal's logic. Then Spreadsheet Analysis. Then the Bloomberg Query Language, to pull data directly instead of clicking through screens. I went from looking up single numbers to writing queries that pulled structured datasets across companies, sectors and years. At some point I realized I wasn't just looking at Bloomberg anymore. I was asking it questions. A terminal stops being intimidating the day it becomes your data source.

One project made it concrete. I built a dataset on AI deployment by industry, with annual data from 1996 to 2025. It covers major US companies and leading global tech firms. Financial fundamentals, R&D, patents, AI investment, ESG, governance, risk. The dataset taught me something I hadn't expected: having access to data is the easy part. Knowing what question to ask is harder. A number by itself tells very little. The same number across time, companies, and sectors can tell a much more interesting story.

That data work fed directly into the research with Dr. Ben Bouheni. A study of how VC backing, ESG and governance affect profitability and valuation, on a 2015-2025 panel of technology and real-estate companies. By the end I was building my own datasets from the terminal and feeding them straight into the regressions. The quantitative methods, OLS regressions and robustness checks, were new to me. But the interesting question was never which regression to run. It was: what can financial data actually tell us about a company, and where do the limits of that data begin?

The final step was the Global Trading Challenge. A virtual $1 million portfolio, long-only, no leverage, maximum 20% per stock. For the first time I wasn't analyzing. I was deciding. The model portfolio split roughly 60% quality growth, 30% momentum and deeptech, 10% event-driven. Analyzing a company is one thing. Deciding what to actually do with that analysis is something else entirely.

The road trips were where the year decompressed. Long drives, no agenda, the country going by. Seeing that much geography in so little time still surprises me when I look back.

Ambition here is social rather than guarded, and that changes how fast things happen. Some of the best memories are the small ones. A Warriors game at Chase Center. An afternoon with friends by the Golden Gate. Watching Roger Federer practice at Stanford. And the graduation ceremony at Menlo, at the end of it all.