Co-founder, Casian · Paris
I'm building Casian, an AI workspace where VC teams screen, diligence and decide on deals against their own thesis, with every claim traced to its source. Before Casian, I was a venture capital analyst in Paris, screening 30 to 40 deals a month for a club of 800+ business angels and its seed fund.
In the funds we talked to, deals live across a CRM, Drive and sourcing tools, and each analyst uses ChatGPT or Claude alone, re-explaining the fund and the deal every time. Casian starts from each fund's own thesis and investment criteria. A team imports the deck and the data room (scanned pages included), screens the deal, runs due diligence theme by theme with web research and a table of evidence, sends questions to the founder, writes the investment memo and votes. The AI proposes and the team decides.
I built the whole product: the web app, the infrastructure and the AI agents. My co-founder Julian Fernandez leads our work with funds. Analysts at VC funds started testing Casian on real deals at the end of September 2026.
In March 2026 I built two DeFi protocols for AI agents, in about a week each. Both run on the Arbitrum Sepolia testnet, with no real capital.
Credit between AI agents and the API vendors they buy from. The agent posts a USDC bond and gets net-30 terms instead of prepaying. If it misses the deadline, the vendor triggers a slash and the bond moves to the vendor's wallet in the same transaction, with no arbitration. The vendor posts a bond too, which it loses if it slashes an agent that paid. Three Solidity contracts (Vault, Registry, Heartbeat) and 131 tests.
One USDC margin vault for trading agents across perpetual DEXs. Each DEX normally holds its own collateral, so an agent with capital on GMX can't use it on gTrade, even when gTrade has the better rate. With UMIP the agent deposits once and the vault routes collateral to adapters for GMX, gTrade and Vertex. Over the same period, the test agent without UMIP hit 4 fragmentation events and the one with UMIP hit none.
Ten interviews with founders, investors and ecosystem actors in the San Francisco Bay Area, from October 2025 to January 2026. An accelerator director summed up the main finding: “never easier to start, never harder to win.” AI is making technical execution cheap, and judgment is becoming the scarce input. The dissertation proposes a Bifurcated Ecosystem Model with six testable propositions.
AI agents can already pay per request (x402) and hire each other for scoped jobs (ERC-8183). A vendor that serves an agent continuously, like an inference API called hundreds of times a minute, still has no way to extend credit and collect when the agent defaults. The thesis argues for a credit and enforcement layer built for agents, and explains why I built AgentEscrow and UMIP.
Aug 2026 - now
AI workspace for VC investment teams. I build the product.
Jul - Aug 2026
I wrote fundraising documents for two startups (€5,000 in total) and coded an internal AI tool that drafted them, cutting each client file from about 20 hours to 2.
Jan - Jul 2026
A club deal of 800+ business angels investing through SPVs, and its seed fund. I screened 30 to 40 deals a month, ran about two due diligences a month with their investment memos, and led the intra-group transfer of the portfolio. The partners also asked me to build the firm's portfolio-monitoring software and workflow automations.
May 2025 - Jun 2026
Ski-Mojo sells exoskeletons for skiers. I worked on the company's €2.85M buyout through closing (financial models, structuring, due diligence) and took part in its expansion into California and Colorado.
May - Sep 2023
Assessed Power-to-X hydrogen projects in South Korea and Japan.
2020 - 2023
Rare-sneaker resale, started at 16. I used private groups that shared live stock by size and by store, ran raffle bots on hundreds of accounts behind my own proxies, and sold in bulk to large French resellers. €90,000 in revenue at a 70% margin.
2022 - 2026
I studied in Paris and did an exchange at ESSEC's Asia-Pacific campus in Singapore. As treasurer of the Student Council in 2023, I managed a €40,000 budget and secured over €350,000 in prizes for events reaching more than 7,000 students. I was also an early member of ESSEC Kryptosphere, Europe's largest student crypto association. My dissertation is in Writing.
Jan - Dec 2025
One of 40 ESSEC students selected for the double degree, which combined an entrepreneurship core with computer science electives. I met my co-founder Julian Fernandez there, and interviewed Bay Area founders and investors for my dissertation between October 2025 and January 2026.
I raced alpine skiing from Paris, eight hours from the Alps, training only on weekends. I ranked first of 175,000 members in France's ESF ranking in 2022 and 2023, raced FIS events, and spent two ski-study terms at Killington Mountain School in Vermont, in 2016 and 2018.