articles
writing on medium. mostly ml systems research and build logs.
caveman tokens: can coding agents say less without becoming useless?july 2026 — benchmarks how far you can compress an agent’s output before it stops being useful→answersurvivalrag: what happens when rag finds the answer, then drops it?july 2026 — measures how often retrieval finds the answer and then loses it while packing context→fastbpe: can we make tokenization faster without changing the tokens?june 2026 — benchmarks bpe tokenizer backends for throughput, latency, and memory at identical output→building a short-term spotify wrapped taught me oauth the hard wayjanuary 2026 — why oauth belongs in the backend, learned by getting it wrong in the client first→how an athlete's amygdala and limbic system react in high-pressure game situationsoctober 2022 — how the amygdala and limbic system drive an athlete’s response under pressure→
04 / 05