Projects

  • Recurrent reinforcement learning trader, reproducing Moody and Saffell's "Learning to Trade via Direct Reinforcement" (2001): a single recurrent neuron that learns to trade by optimizing the Sharpe ratio directly.

    Jupyter Notebook
  • A tool that analyzes discrete math proofs written in LaTeX and parses natural language to identify common proof type-errors.

    OCaml
  • A game where you navigate a maze of Caltech buildings, avoiding seekers at every turn. Outsmart them, hide cleverly, and reach your target unscathed.

    C
  • Vision Transformer for MNIST-in-the-wild dataset created by EE/CS148a 2026 students. MNIST-in-the-wild is a small, human-interpretable dataset of in-the-wild numerical digits designed to be challenging for machine learning models.

    PythonReport
  • A language model implemented from scratch.

    Python
  • Score-based diffusion and rectified flow trained on FashionMNIST.

    Python
  • 3D Operators, 3D Data structures, Silhouette rendering, Textured rendering.

    PythonReport
  • Project covers Time Integration, Variational and Symplectic Time Integrators, The Discrete Lagrangian, The Discrete Euler-Lagrange Equations, Keyframe Interpolation, Splines, B-splines and NURBS.

    C++
  • Parallelization of Hidden Markov Model (HMM) algorithms.

    PythonReport
  • Language model benchmarks, profilers and performance optimizations.

    Python
  • A vision language model built (mostly) from scratch.

    Python
  • A zero-config open-source tool for documenting Node.js Express API endpoints with Swagger.

    TypeScript
  • Fast and minimal ray tracing with custom rendering and lighting implemented in Rust.

    Rust
  • High-availability infrastructure and automatic failover cloud architecture.

    Shell
  • SQL queries for occupancy calculation, optimal room suggestions, hotel bookings ARI and geolocation.

    SQL

© 2026 by Divin. Design inspired by Goodfire