Writing
Notes from building AI systems.
Essays and guides on AI systems and evaluation, drawn from the decisions and failures in my projects.
Filter
2026
- Brief Essay~1 min readA validator is not a better promptWhy DBWhisper gates every generated query with deterministic code instead of a stricter system prompt — 'less often' is not a safety property.
- Essay~3 min readTrust is not a safety modelWhy I gate an LLM's SQL behind a deterministic, fail-closed validator instead of trusting the model to behave.
- Essay~2 min readLook-ahead-free by constructionThe most common way a backtest lies to you is look-ahead bias. Here's how I designed it out of TradePulse instead of hoping to catch it later.
- Guide~2 min readRetrieval beats stuffing the schemaA practical approach to natural-language-to-SQL on real databases: document the schema, embed it, and retrieve per question instead of pasting the whole thing into the prompt.
New writing, occasionally.
The odd essay or note when I ship something worth reading — no cadence, no spam, unsubscribe in one click.