Projects

Projects

Code I’ve written in public, newest first. Every project links to its GitHub repository.

LLM and inference work

pytorch_tutorials: building an LLM from scratch

My working notebook for Sebastian Raschka’s Build a Large Language Model (From Scratch), plus PyTorch fundamentals. So far it covers:

  • a simple tokenizer and BPE with tiktoken
  • a sliding-window GPT dataset, plus token and positional embeddings
  • self-attention, then causal attention, then multi-head attention
  • a full GPTModel with LayerNorm, feed-forward blocks and weight tying, and autoregressive text generation
  • CPU vs Apple Silicon (MPS) benchmarks and a Distributed Data Parallel (DDP) training example

Write-ups: The Flashlight and the Shadow: dot products and attention.

jev-tutorials: calibrated decision models for support triage

A Node.js, Express and TypeScript app (Vercel AI SDK) that demonstrates TypeSafe AI Jev’s non-generative decision primitives (Choice, Score and Boolean). It runs them in a streaming retail-support ticket pipeline with a policy engine.

Coming next: GPU kernels lab

Triton and CUDA kernels written while I work through Vizuara AI’s AI Kernel Engineering program: fused softmax, tiled matmul, and a minimal FlashAttention forward pass, each benchmarked against PyTorch. (Repository not published yet.)

Infrastructure and MLOps

project-ml-microservice-kubernetes: ML inference API on Kubernetes

Takes a pre-trained scikit-learn model served by Flask and operationalizes it. The steps are linting, a Docker image, a Kubernetes deployment, structured logging and CircleCI. Built for the Udacity Cloud DevOps Engineer Nanodegree.

cloud-devops-project

Project from the Udacity Cloud DevOps Engineer Nanodegree.

Earlier machine learning projects

  • FER: facial emotion recognition with heuristic and transfer learning (write-up).
  • Airbnb: price prediction for Airbnb listings.
  • SparkGen: a React tool that generates Apache Spark job configurations, built during my DataSpark internship.
  • Altair-Viz: companion code for my Altair visualization tutorial (write-up).

Tools

I’ve also built a set of personal finance calculators in React and TypeScript.