Articles
Technical writings on data science, AI evaluation, simulation, and logistics optimization.
Being 200 cm tall is supposed to make you fast. The data tells a more complicated story: longer strides and lower cadence, yes — but also a heavier body to carry over every kilometre. Analysis of 767 personal runs against the Ferber et al. dataset of 739 runners.
Read Article →
A running log of recurring failure patterns when using LLMs for coding — over-engineering, clumsy solutions, token budget surprises, and more. Each pattern includes when it was first observed and whether it's still present. Extended over time as new patterns emerge.
Read Article →
A personal history of code completion — from teaching myself QBasic in the 1990s and IntelliJ IDEA in 2001, through GitHub Copilot's leap to intent-based suggestions, to today's agentic coding with Claude Code. What actually changed, and what it means for the developer's role.
Read Article →
Building an evaluation framework for LLM agents at Meight. When extracting structured shipping data from documents, we learned that evaluation requires both strict metrics (for production readiness) and LLM-as-a-judge (for semantic correctness).
Read Article →
An experimental evaluation of how 8 models (6 cloud, 2 local) perform on logic puzzles, revealing the gap between pattern matching and first-principles reasoning. Includes complete raw model responses.
🎯 Try Interactive Challenge → Read Full Analysis → 📝 Raw Outputs →
Deep dive into building a large-scale discrete event simulation system for Glovo's global delivery network, covering architecture decisions, performance optimization, and real-world validation.
Read on Medium →