Day 5: Microsoft Hackathon — Model Grading, Parameter Friction, and Next Questions
Pipeline runs, thread summaries graded, and model friction surfaced. Next test: can insights drive action?
Pipeline runs, thread summaries graded, and model friction surfaced. Next test: can insights drive action?
Debugged a broken email threading approach, replaced it with a simple recursive pass, and resumed grading LLM-generated summaries.
Filtered 424 threads to ~250 candidates, added package/entity tagging, and built an agent writing editing workflow.
A whimsical moment observing a retired couple’s daily ritual and a surprise silly hat.
Distractions derailed metadata work; brainstormed importance heuristics and moved the MVP to SQLite for scale.
Hackathon MVP: building a lightweight LLM-driven summarizer for Debian mailing lists — architecture, data collection, and early lessons.
A reality check on ‘#EndOf10’ panic: numbers, the used laptop market, and practical ways to help users migrate safely.
Notes and key takeaways: the Theory of Constraints, five focusing steps, and practical change questions.
A practical, non-technical tour of SPF, DKIM, DMARC and ARC — why modern email needs DNS help and what to do about it.
A dad-and-daughter twist on the NYT Strands game — play the custom ‘Stella’ puzzle and share the results.