Seraphim Labs · Building

Software for the
limits of human performance.

Computer vision and applied ML for movement, form, and feedback. Currently building FormLab AI at Seraphim Labs.

  • NowFormLab AI
  • ProofKaggle Silver
  • ResearchUPenn · CML
  • BasedPhiladelphia
Currently building

FormLab AI.

A vision system that watches how you move and tells you what to fix.

I'm building the feedback loop that's been missing from physical practice: millisecond-accurate pose analysis, biomechanical reasoning, and coaching that adapts to the individual instead of the average.

The hard parts are real-time inference on consumer hardware, modeling form as a structured signal rather than a classification, and earning trust from athletes and clinicians. I'm treating the body the way modern tooling treats code — instrumented, diff-able, improvable.

  • 01

    Real-time perception

    Pose and motion estimation tuned for sub-frame stability on commodity cameras.

  • 02

    Structured form

    Biomechanics-aware models that reason over joints and trajectories, not just labels.

  • 03

    Adaptive coaching

    Feedback that learns the individual — range, constraints, goal.

Technical proof

Kaggle Silver Medal.

Top-percentile finish in LLM 20 Questions, an open Kaggle competition against a global field of researchers and practitioners.

Won by getting the modeling, validation, and engineering details right — not by scale. The kind of result that comes from caring about the tenth decimal place and the shape of the data, not just the headline metric.

Result
Silver Medal
Competition
LLM 20 Questions
Discipline
Applied ML & modeling
Research

Computational Memory Lab, UPenn.

Applied ML and systems work under Dr. Michael J. Kahana.

Built an automated annotation framework for continuous EEG and speech, pushing onset detection toward sub-millisecond precision with WhisperX and adaptive thresholding. Engineered closed-loop Unity experiments that deliver stimuli at EEG-predicted moments of high memory encoding.

Real research code, real subjects, real timing budgets.

  • EEG
  • WhisperX
  • Closed-loop systems
  • Unity / C#
  • Signal processing
Selected

Work.

  • Previously

    Meta

    Software Engineering

    Past
  • Incoming

    Google

    Software Engineer Intern · Search

    Summer 2026
  • Incoming

    Suno

    Software Engineer Intern

    Fall 2026
Thesis

The interesting problems left are the ones where models meet the physical world and have to be right. I build at that boundary — vision, signal, feedback — and I ship.

— James

Let's talk.

Open to investors, technical collaborators, and operators who care about this problem space.