Building this site

This portfolio renders from a single resume.json. The goal: a new resume is parsed, validated with zod and lands as a reviewed pull request before Cloud Build deploys it to Cloud Run. Validation and rendering work today; the pull request and deploy stages are in progress.

Status: In progress
Dates: Oct 2026
Tags: Next.js
TypeScript
zod
MDX
Cloud Run

Problem

Keeping a portfolio in sync with a resume by hand drifts. This site renders all career content from one validated file, so the resume is the source of truth.

TODO(saketh): in your words, why you built this and who it is for.

My role

TODO(saketh): your role (e.g. sole designer and engineer) and how you used Claude Code with guardrails.

Architecture

resume.pdfinbox
Parsezod
resume.jsonvalidated
PR reviewhuman gate
Cloud Rundeploy

Fig. 2 · Resume-to-site pipeline (PR review and deploy in progress)

Drawn: SK
Rev: A
Status: In progress

Content is data: content/resume.json and content/projects/*.mdx are validated with zod at build time, and an invalid file fails the build. Project frontmatter uses a strict schema:

export const projectFrontmatterSchema = z.strictObject({
  slug: z.string().regex(/^[a-z0-9]+(-[a-z0-9]+)*$/),
  title: text,
  summary: text,
  featured: z.boolean().optional(),
  tags: z.array(text),
  status: z.enum(["in-progress", "shipped", "archived"]),
  diagram: diagramSchema,
});

Key decisions and trade-offs

Cloud Run over Vercel. TODO(saketh): the reason in your words (PROGRESS.md records it as "for the infra talking point").
Strict zod objects as the PII guard: unknown keys such as phone fail validation, so a resume parse that carries them over cannot be published.
WCAG 2.2 AA as a test gate: a contrast matrix over the design tokens in both themes, plus axe-core on every route.
MDX is evaluated at build time only; no MDX runtime ships to the browser.

Outcome

TODO(saketh): the outcome. No metrics until real ones exist.

What I'd change

TODO(saketh): what you would do differently.