Notes from systems that had to work.
This is not a blog. It is the working reference our own engineers use — the performance problems we have actually hit in production, what caused them, and what fixed them. We publish it because a firm’s standards say more about how it builds than any case study can.
Start here
- 01Threat modelling a public-sector systemTrust boundaries, the abuse cases that actually occur, and what each costs to stop
- 02Getting cited by answer enginesWhat decides whether an assistant quotes your page — and how to measure it
- 03Grounded assistants that refuse wellRetrieval, guardrails and refusal design, including the bugs we hit building ours
Browse by area
Security & assurance
Threat modelling, audit trails and the evidence an institutional assessor asks for.
2 articles02Search & discoverability
Technical SEO on modern React stacks, and being cited by the answer engines buyers now start with.
2 articles03Applied AI
Evaluating models honestly, and putting them in front of people without inventing anything.
2 articles04Performance
Latency budgets, PostgreSQL under load, and measuring what users on mid-range devices actually experience.
11 articlesThese notes come from the same systems described in our engagement record — a state legislature, live payment rails, field operations at scale.
See the work