Skip to content

Writing

Notes from the work

What broke, what it cost to find, and what we changed so it could not happen the same way twice.

31 documents · press / to search

  1. 7 min read

    The cost of JavaScript: what a 100/100 Lighthouse score actually requires

    A framework floor you cannot budget below, four techniques that move the number, and the measurement discipline that keeps it there. Written from a site that scores 100/100/100/100 with zero route-level JavaScript.

    PerformanceCore Web VitalsFrontendNext.js

  2. 6 min read

    Bridging Figma and code: a design system built on tokens and Tailwind

    Three token layers, one configuration knob for a whole colour ramp, and contrast ratios computed by a test rather than typed into a comment — how to build a design system where the code and the design file cannot drift apart.

    Design systemsTailwindFrontendAccessibility

  3. 7 min read

    Web accessibility as a competitive advantage: an engineering checklist

    The accessibility defects an automated suite finds, the ones it structurally cannot, and eleven real failures from one build — what caused each, and the check that now prevents it.

    AccessibilityWCAGFrontendTesting

  4. 7 min read

    Beyond the wrapper: building resilient multi-agent AI workflows in production

    What separates a demo from a system that runs unattended — a policy boundary written before anything executes, idempotent tools, bounded loops, and a failure model that assumes the model will be wrong.

    AIAgentsArchitectureBackend

  5. 7 min read

    Building a production RAG pipeline: the latency, cost and accuracy trade-offs

    Every decision in a retrieval pipeline trades one of three things against the other two. Here is where each trade sits — chunking, hybrid retrieval, reranking, caching — and how to tell which one your system is actually losing.

    AIRAGVector searchArchitecture

  6. 6 min read

    PostgreSQL with pgvector or a dedicated vector database: how to choose

    A decision guide rather than a benchmark — what each option actually costs you in operations, consistency and query power, the scale at which the answer flips, and the honest test to run on your own data.

    Vector searchPostgreSQLAIArchitecture

Start here

Tell us what you are building

Or what is breaking, or what has to go faster. You will get a straight answer from an engineer who would do the work.

Start a project