I'm a machine learning engineer who likes to build AI/ML models all the way to production! Most of my time goes to Polycreek, a child-safety nonprofit I founded and lead, where I built and run a grooming-detection classifier that serves partner platforms in production.

What I work on

• Fine-tuning transformer models (SFT, LoRA/QLoRA) and shipping them: quantized inference, scale-to-zero GPU serving, eval harnesses, and release gates that actually block bad models.
• At Polycreek the production classifier is a ModernBERT backbone with seven classification heads, running at 0.98 recall and a 0.011 false-positive rate on roughly 10k requests a day.
• Contract work as a SME on model architecture, evaluation and post-training data pipelines for frontier AI labs.

Research I published an empirical study on why overparameterized networks generalize: Implicit Regularization and Generalization in Overparameterized Neural Networks (arXiv:2604.07603). It covers SGD batch-size effects, flat vs. sharp minima via Hessian analysis, double descent, the neural tangent kernel, and the Lottery Ticket Hypothesis on CIFAR-10 and MNIST. Code and figures are here: overparameterized-generalization.

I have a few other ongoing research projects in the fields of RSI, Emergence and Agentic Failure Prediction which are soon to be published in late 2026 to early 2027

A few things I've built for fun

• overparameterized-generalization - the experiments behind the paper above
• Datapack - a universal TypeScript data toolkit that allows schema to be defined once and used wherever
• SAT-Solver-Project - a from-scratch SAT solver (DPLL + CDCL + VSIDS) with clause learning and visualizations
• mini-ppl-0106 - a minimal probabilistic programming language with MCMC inference
• DFA-Hopcroft-s-Algorithm-Assessment - DFA minimization via Hopcroft's algorithm

Toolbox: Python, Rust, PyTorch, HF Transformers, FAISS/pgvector, Docker, AWS, Terraform and more that I don't think anyone would care to read!

Elsewhere: polycreek.org · LinkedIn
Always open to conversations with likeminded engineers!