Hidden Jobs Engine

A static, client-side job search engine for senior AI-infrastructure roles: inference / model serving, GPU clusters and AI data centers, ML platforms, and the Kubernetes / DevOps / SRE work underneath them. Built with Astro; no backend, no API keys, no tracking.

Live at https://shubhamtatvamasi.com/jobs/

Develop

npm install
npm run dev       # http://localhost:4321/jobs/
npm test          # engine unit tests (node --test)
npm run build     # static site → dist/
npm run preview

Deploy

Every push to main runs .github/workflows/deploy.yml, which tests, builds and publishes dist/ to GitHub Pages (repo Settings → Pages → Source: GitHub Actions). The site is served under the /jobs base path of the shubhamtatvamasi.com user site — see astro.config.mjs.

What it does

• Hidden-jobs scan — fetches live openings straight from the public JSON APIs of 56 AI-infrastructure employers (Greenhouse, Ashby, Lever), grouped as GPU clouds & AI data centers, inference platforms, frontier labs, AI silicon, and ML / cloud-native platforms. Ashby boards include real posted salary bands.
• Aggregators — RemoteOK, Remotive and Jobicy APIs, merged and deduped.
• HN "Who is hiring?" — pulls the latest monthly thread via the Algolia API and filters founder-direct posts to your role.
• Deep links — pre-built boolean queries for Google/LinkedIn plus ATS X-ray searches (site:job-boards.greenhouse.io etc.).
• Filters — role family (inference, GPU/HPC, AI/ML platform, data center, Kubernetes, platform/DevOps/SRE), seniority, remote scope (🌍 global / 🇮🇳 India-eligible / 🇺🇸 US), salary floor and sorting. The default seniority, Staff+ or unlevelled, keeps the "Member of Technical Staff" and level-less titles AI labs and startups use. "Best match" ranks inference/GPU titles above generic platform ones.

Sections are hash-routed (#scan, #agg, #hn, #links, #companies, #guide), so any tab can be linked directly. Light/dark theme follows the OS and can be toggled from the header.

Layout

Path	Purpose
src/pages/index.astro	page shell; server-renders controls, companies, deep links, playbook
src/components/JobPanel.astro	shared markup for the three live-fetch sections
src/layouts/Base.astro	<head>, fonts, theme bootstrap
src/styles/global.css	design tokens + layout
src/lib/data.js	verified companies, endpoints, deep links, role/skill tuning
src/lib/engine.js	pure logic: API adapters, salary parsing, remote classification, scoring
src/scripts/app.js	client glue: tabs, filters, fetch orchestration, rendering
test/engine.test.js	Node tests for engine.js

All endpoints, company slugs and deep links verified live on 2026-10-10 (56 scannable employers; NVIDIA, AMD, Groq, Hugging Face, Mistral and the hyperscalers use Workday/Gem/Workable/custom sites and are listed for manual checks).

Remote classification notes:

• Ashby marks hybrid postings isRemote: true; the adapter trusts workplaceType instead, so office-hybrid roles no longer show as remote.
• Greenhouse has no structured remote flag: a location spanning several cities/regions ("Central - United States", "Austin; Boston; ... United States") is read as remote-eligible-in-region; single-office listings ("San Jose, California, USA") stay on-site.