Case studies

A collection of deep-dives, stats, and random technical rabbit holes I fell down. Most of these were written at 3 AM fueled by a bunch of dynamic code issues, deep existential questions, and an incredibly unhealthy amount of late-night coffee.

  1. Deterministic Inference for LLMsResearchA deep dive into floating-point non-determinism in parallel GPU architectures, mathematical PPO variance in RLHF updates, and batch invariance engineering.—Research
  2. Abandonment IssuesResearchA statistical diagnosis of dropout patterns in digital forms using regression models and NLP sentiment analysis.2024Research