Writing

Notes from building.

Decisions, observations, and open questions from my personal AI projects. Each note links back to its evidence.

Field notes

What the experiments taught me.

RAG evaluation · 3 min read

What ten RAG queries can and cannot tell you

Reading a small evaluation without turning an average score into a claim of production readiness.

Read the note Agentic systems · 2 min read

Human approval is a product feature

What a reviewer needs to see between a model-generated signal and an action.

Read the note Multilingual fine-tuning · 3 min read

More English terms do not prove a better answer

What changed in four Arabic-English model comparisons, and what the metric leaves unanswered.