Writing
Notes on AI engineering, agents, RAG, and building products that hold up in production.
Interview prep is full of question banks and resume puffery. Almost nothing simulates the actual conversation. Here's the gap I kept seeing — and the product I shipped to close it.
If the only way you know your AI feature works is that it felt right in the demo, you do not actually know it works. Here is how to start measuring.
The skill that matters is not wording a clever prompt. It is deciding what goes in the context window, what stays out, and in what order.
Three levers that actually moved our token spend: caching what repeats, routing to the right size of model, and sending less in the first place.
Stuffing an LLM's whole memory into every prompt is the most expensive habit in agent engineering. Here is how the memory crowd is fixing it, and the four-way retrieval I keep reaching for.
The honest answer nobody wants to give you ,yes, no, and it depends entirely on what you do next.
A short introduction and what to expect from this blog.