Blog & Notes
Notes on AI research, intelligent systems, and the engineering behind the projects I build.
Learn to separate user intent, tool responses, observed operations and remote outcomes, with a small approval checker, failure scenarios and a practical evaluation plan.
A practical guide to neural interpretation, explicit constraints and checked decisions: build a small validator, understand specification gaps, and design an evaluation that separates prediction from execution.
Understand directional factual recall, distinguish parameter learning from in-context reasoning, build paired probes, and evaluate bidirectional retrieval without overstating the original research.
An engineering walkthrough of authorization, state transitions, reconnect recovery, document access and bounded AI drafting, with concrete failure scenarios and an honest evaluation plan.
Follow ingredient normalization and cosine retrieval, uncover a filtered-index mismatch in the current code, and work through a corrected candidate-search example and practical evaluation plan.
Learn how to separate sensing, vision, control and delivery; handle stale data and placeholder measurements; and design bounded buffers, meaningful benchmarks and failure-injection tests.