Technical publication
Writing that shows the work.
Experiments, field guides, and essays about how AI systems behave, where evaluations break, and what the results mean in practice.
9 entries in the current archiveMy Agent Signed Its Own Timesheet
An AI agent logged a tool call it never made. The fix was not a better instruction—it was a mechanical boundary with an independent record.
The Mistake My Network Refused to Fix
A neural-network debugging investigation into a model that kept repeating the same error, even after the obvious fix.
The Fast Algorithm That Was 370 Times Slower
An empirical look at the difference between theoretical speed and real performance when implementation details take over.
When Washington Stopped Being a Person
A text-analysis investigation into how language, representation, and context change what a model believes a name means.
VerifAI: Teaching an AI to Check Its Sources in Two Languages
How I built and evaluated a bilingual, evidence-grounded question-answering system designed to distinguish retrieval from invention.
When AI Learns to Doubt Itself, Medicine Gets Safer
Why uncertainty is not a weakness in clinical AI, but a necessary signal for safer human judgment.
What Movies Mean to People: Content Embeddings vs. Behavioral Embeddings
A hands-on experiment comparing collaborative filtering and content-based embeddings on MovieLens-100k, and what the results reveal about how taste actually works.
Democracy in Data: What an Unsupervised Algorithm Found and Why It Matters Now
What unsupervised learning reveals about the structure of democracy, and where quantitative patterns need political context.
Machine Learning Algorithms Every Data Scientist Must Know
A visual field guide to model families, learning paradigms, and the decisions that connect them.