Home / Blog
Engineering notes
Engineering notes.
What we learn building agentic systems, voice agents, retrieval, and automation for real workloads — written up so you can steal it.
What it costs to run an AI feature in production
A running cost model is built from tokens per completed task and tasks per user per day, rather than from a provider headline rate.
Prompting, retrieval or fine-tuning: how to decide
The three options change different components of a system, so the right one follows from what is actually failing.
The latency budget of a production voice AI agent
A millisecond-by-millisecond breakdown of where response time goes in an AI phone agent — and the engineering that claws it back.
Why naive RAG fails in production (and what to build instead)
The tutorial pipeline works on clean text and collapses on real documents. Where it breaks, and the stack that doesn't.
Designing guardrails for autonomous AI agents
Autonomy without guardrails is a liability. The engineering patterns that let an agent act on your systems without keeping you up at night.