Max De Jong

I am an applied scientist broadly interested in unlocking reinforcement learning, improving sample efficiency, and robustly learning from imperfect labels. I have spent the past five years solving machine learning problems at the boundary of research and engineering, mostly at small startups. I am trained as a theoretical biophysicist, where my PhD research advanced our understanding of the emergence of antibiotic resistance in microbial communities by combining mathematical modeling, high-performance computing, and machine learning algorithms.

Portrait of Max De Jong

Recent writing

All writing →

Active ideas

These are the projects that I am actively working on today as I have time:

  • Small Language Models + RL Investigating task-specific small language models optimized using reinforcement learning on a single consumer graphics card.

  • Agentic RL Exploring self-evolving agentic frameworks using role-specific small language models with verifiable rewards.

  • 3D Pose Estimation Applying the latest computer vision models to robustly extract biomechanical metrics to analyze powerlifting form.

Timeline

  1. Harpin AI

    Combined agentic AI, predictive models, and identity resolution to understand customers.

  2. Agentic AI

    Trained neural networks to play video games.

  3. FICO

    Applied explainability methods to financial models.

  4. University of Michigan

    Completed biophysics PhD studying emergent behavior in mathematical models of antibiotic resistance.

  5. Caltech

    Completed undergrad studies in applied physics.