A résumé you can talk to

An open web front for resume-agent: a calm, single-page conversation thread where a visitor asks questions in plain language, scores a job's fit, and generates a tailored résumé — and where an LLM or agent can read the whole profile straight from the page. The reference instance runs at yuens.me.

Nothing here hardcodes a person. The site queries a profile (/info) and renders it; point it at your own resume-agent and it's your site.

The other half — the truth layer. This is just the front. The profile data, the query/agent API, and the honesty contract live in resume-agent — also open source, and a simple, vanilla backend you can stand up as-is. This portfolio works with it from day one: build your truth there, and this surfaces it to humans and machines. Two repos, one queryable you.

Why this exists

I built this for myself — a résumé I could keep honest. A résumé is usually the most static, hand-tuned, slightly-dishonest document a person owns, and it goes stale the day you stop editing it. I wanted the opposite: a profile that's queryable, versioned, and stays true to itself without manual tuning — one an AI can read fairly, because AI is increasingly the first reader.

Building it for myself is what made it open. None of it hardcodes me; the engineering is the same for anyone. So if it's useful, take it. If a few people get seen a little more fairly because of it, that's the difference, and it compounds past me.

The engineering is the portfolio — it doesn't describe how I work, it demonstrates it, live, to anyone (human or model) who looks.

(— draft, distilled from the build conversation. Edit it to your truth.)

Lineage

The grandfather of this idea is OB1 / Open Brain by Nate B. Jones — infrastructure for your thinking: one database, one AI gateway, one channel, any AI plugs in. It's inward by design — a private place to reason, with no outward reach. It never tries to speak those thoughts in public, or to let your actions be the thing others trust. That's its honest scope.

The outward half is what resume-agent and this front add: they take the introspection layer and give it a witness — a queryable, public record. Because a truth kept to yourself is only half a truth; it has no outside witness to validate it. Publishing it, and letting humans and machines query it, is what makes it true. Trust by action, in the open. The grandfather gave the mind a place to think; this gives it a place to be seen.

Architecture

Browser ── direct ──▶ agent.yuens.me   /info  /query  /match        (public)
Browser ── /api/resume ──▶ this server ── Bearer key ──▶ agent.yuens.me/resume (SSE)
LLM / crawler ──▶ this server  →  JSON-LD · <noscript> profile · /llms.txt · /robots.txt

• Human surface: a React + Vite SPA. /info → profile, /query → answers, /match → fit score, called straight from the browser (CORS-open). /resume is key-gated, so the tiny Hono server (server/index.ts) proxies it with RESUME_AGENT_API_KEY server-side — the key never ships to the client.
• Machine surface: the server makes the site self-describing for non-JS consumers — schema.org/Person JSON-LD, a crawlable <noscript> profile, discovery links, real /robots.txt + /llms.txt, and /.well-known/agent-card.json → backend. All derived from the live /info (cached, refreshed every 10 min) — see server/machine-surface.ts.

Stack: React 19 · Vite · TypeScript · Hono · react-markdown · qrcode.react. Design system is the hand-authored src/styles.css. Deploys to Railway (or any Node host).

Fork it — your own queryable résumé

1. Deploy your own resume-agent — it holds your profile data and the query/agent API.
2. Fork this repo and point it at your backend + site (see Environment): VITE_API_BASE + RESUME_API_BASE = your resume-agent base · VITE_SITE_URL = your canonical URL · RESUME_AGENT_API_KEY = your resume-agent key (server secret).
3. Swap the favicons in public/ (see Customize).
4. Build & deploy (below).

Your name, role, summary, skills, projects — and the JSON-LD / llms.txt / <noscript> machine surface — all come from your /info. The published <title>, description, and canonical are derived from your profile + VITE_SITE_URL; no identity is baked into the code.

Develop

npm install
cp .env.example .env      # set the vars below (RESUME_AGENT_API_KEY for the résumé proxy)
npm run dev               # vite on :5000 + tsx server on :8787 (/api proxied)

Append ?instant=1 to skip the typewriter/entrance animations. The machine surface (JSON-LD / <noscript> / llms.txt) only renders in the production server, so test it with npm run build && npm start.

Build & run

npm run build             # vite build → dist/  (needs VITE_API_BASE + VITE_SITE_URL)
npm start                 # tsx server/index.ts — serves dist/ + /api/resume + machine surface

Environment

Var	Scope	Notes
VITE_API_BASE	client (build-time, public)	your resume-agent base, e.g. https://agent.yourname.dev
VITE_SITE_URL	client (build-time, public)	your canonical URL; drives canonical/og:url + the agent-menu QR
RESUME_API_BASE	server	backend the /api/resume proxy + machine surface read from
RESUME_AGENT_API_KEY	server (secret)	your resume-agent API_KEY; never committed, never in the bundle
PORT	server	host injects this (Railway etc.)

Customize

• Identity (name, role, summary, skills, projects, contact): edit your profile in resume-agent — it flows through here automatically, no code change.
• Favicons: public/favicon.ico, favicon-16.png, favicon-32.png, apple-touch-icon.png.
• Copy & samples: greeting headline (src/components/Greeting.tsx), sample job posts (src/components/MatchResume.tsx), accent color (src/styles.css → --accent).
• Machine surface: server/machine-surface.ts (JSON-LD, llms.txt, robots.txt).

Deploy (Railway)

Build npm run build, start npm start, healthcheck /healthz. Set the env vars on the service (RESUME_AGENT_API_KEY as a secret). Point your apex/www domain at it; the resume-agent backend stays on its own domain.

License

MIT — © 2026 Sunny Yuen. Fork it, make it yours, keep it honest.