Writing outside the papers — on agentic systems and knowledge design, and the occasional data-driven analysis from a physicist's angle.

Subscribe via RSS

• Cutting agent cost

  Cascade-a reduced measured coding-task cost by 70.5% on 13 golden tasks, with the crucial limit that escalation was never exercised.

  2026 · 3 min read

  • ai-engineering
  • llm
  • agentic
  • testing

• Price/performance agent workloads

  Measuring agent jobs exposed an unreliable summary judge; pairwise comparisons and deterministic checks give a more careful basis for model choices.

  2026 · 6 min read

  • ai-engineering
  • llm
  • agentic
  • testing

• What Legacy Code Taught Me About Learning AI From Scratch

  LinkedIn post on what CERN legacy code taught me about durability, and building a transformer from scratch to learn AI fundamentals.

  2026-07-05 · 2 min read

  • ai-engineering
  • machine-learning
  • pytorch
  • cern
  • learning

• LinkedIn post - AI should navigate knowledge, not rediscover it

  Published post on the agentic knowledge-base approach and context discipline.

  2026-06-09 · 1 min read

  • ai-engineering
  • agentic
  • knowledge-base

• COVID-19 super-spreaders - a data-driven analysis (2020)

  Public analysis arguing super-spreader events break SEIR models; a physicist's data-driven take.

  2020-04-23 · 1 min read

  • data-analysis
  • statistics
  • physics