Latest Post
Lunchbench: Five Ways to Fail a Lunch Audit
Fourteen models, from cheap open weights to frontier, ran the first real task in my personal AI benchmark, a read-only audit of a daycare lunch calendar. The ranking mattered less than the five …
Recent Posts
- Lunchbench: Five Ways to Fail a Lunch Audit Aug 12, 2026
- Building a Personal AI Benchmark Jul 28, 2026
- Building a Local OCR Pipeline for LLM Document Understanding May 3, 2026
- Teaching TinyLlama to Refuse Coffee: My First Fine-Tuning Experiment Sep 25, 2025
- Building a Neural Network Visualizer: Lessons from a 111k Parameter MNIST Model Sep 13, 2025
- Understanding Self-Attention: A Hands-On Exploration Jul 20, 2025
Projects
pdb-mutator
Kubernetes admission controller that automatically fixes PodDisruptionBudget configurations
View on GitHub →Neural Network Visualizer
Interactive visualization tool for understanding neural network architectures and MNIST digit classification
View on GitHub →