sorawit.com
Sorawit Wanitwarodom · Web & design systems

Trust is a property of how a system is built, not a claim on top.

I build interfaces where the craft lives in the details no one points at — honest loading states, deliberate pacing, failure modes you can actually read.

static Figma responsive + motion designer review approved

How I work: I take a static desktop comp and make it real — responsive, in motion — then refine with the designer until it’s right.

  1. SvelteKitElysia / BunSSE real-timeAIi18n

    Declair

    A decision-clarity platform: pairwise comparison voting with Condorcet aggregation, real-time results over server-sent events, and AI-assisted poll creation. Designed, built, and shipped end-to-end — solo, across 20 locales.

    The signal: Ships a real product alone, with craft — from real-time architecture to internationalization across 20 locales.

    declair.ai
  2. Agent governanceGATE→WEIGH→VERDICTEvalsTypeScript

    Kerby

    A governance harness for AI agents — a GATE→WEIGH→VERDICT loop that gates a risky action, weighs the call, and records a verdict you can audit. Spun out of the skills shelf into its own versioned system.

    The signal: the site's thesis, made executable — trust enforced at the gate, not asserted on top.

    kerby
  3. Claude CodeMCPSub-agent orchestrationEval designTypeScriptThai NLP

    Agent Skills

    agent-skills

    A shelf of Claude Code skills: creative tooling, content workflows, AI-product review, sub-agent orchestration. An exploration of harness engineering — designing the context, feedback loops, and eval discipline around agents.

    agent-skills
    konthai

    A standalone skill in its own repo: decodes deliberately obfuscated Thai into standard Thai — สก๊อย glyph-substitution, RO-leet glyph-art names, phonetic respelling. Built from a verified glyph reference, and it flags spans it can't resolve instead of guessing.

    konthai

    The signal: builds leverage-multiplying tooling, and reasons about agents in the terms the hard problems demand — harness engineering, evals, context design.