Building AI agents that manage real money. Started with ML in drones and submarines, ended up teaching machines to beat humans at DeFi.

ARMA manages $11M+ in DeFi yield optimization. It's the first AI agent that's actually profitable instead of just burning through VC money. Five months in and it consistently outperforms manual strategies across multiple protocols.

Leading a team of 12 engineers. Raised +$5M from CoinFund, StarkWare, and Arrington Capital.

UPM → Defense → BBVA → adidas → Giza

Started with AI for autonomous systems (drones/submarines) in EU defense research projects. Moved to banking as Head of ML at BBVA creating their whole production-grade infra, then scaled consumer ML with recommender system at adidas for millions of users daily.

ARMA (2025)
$11M+ AUM autonomous yield optimization agent. First profitable AI agent in crypto.

Orion ZKML Framework (2023-2024)
Open-source framework for verifiable ML on-chain using STARKs. 35+ contributors.

Giza Platform (2023-2024)
Production ZKML cloud serving DeFi protocols.

adidas MLOps (2022-2023)
ML platform handling millions of daily requests globally. Reduced AI product time-to-market by 60%, optimized preprocessing/training times by 80%. Built feature store scaling to 21k TPS for real-time recommender systems.

BBVA ML (Previous)
Led ML infra for banking operations. Financial ML and risk models.

Hackaton Awards

• 🥇 Best Protocol Hack - ETHGlobal NFTHack
• 🥇 Zorb's Hack Choice - ETHGlobal NFTHack

def learned_the_hard_way():
    return {
        "hiring": "Hire engineers who debug production, not just tune hyperparameters",
        "product": "Profitable AI beats perfect AI every time",
        "scaling": "Build systems that work when you're asleep", 
        "strategy": "Open source the framework, monetize the infrastructure"
    }

Most yield optimization is manual and emotional. ARMA uses continuous mathematical optimization. The difference comes from applying lessons from regulated banking ML and consumer-scale systems to an industry that needed better engineering discipline with lack of personalization.

Currently focused on expanding ARMA to more chains and launching institutional programs. The goal is proving AI agents can be trusted with serious capital when built properly.

If you're working on production ML systems that handle real value, I'd love to connect.

🤖 Build agents that work • 👥 Lead teams that ship • 💰 Manage real capital