Story

I built a pipeline that fetches all HN job postings (both the monthly "Who is Hiring?" threads and direct job posts (/jobs)) classifies each one using an LLM (Gemini Flash via OpenRouter), and publishes the results as a static site.

What it tracks: - Top technologies and roles with month-over-month changes - Remote work breakdown (fully remote, hybrid, onsite) - Salary and equity mention rates - Experience level distribution - Per-job classification (click a technology to see its role breakdown, or vice versa)

Jobs are sent to the LLM in batches of 10 with structured output (JSON Schema). Total cost: ~$0.07/run, under $2/year. The pipeline writes JSON to a GitHub repo served via Pages. The site is a static Next.js export that fetches the JSON at build time. No server, no database, no ongoing costs beyond GitHub Actions minutes (free tier).

Everything is open source: - Data pipeline: https://github.com/gantryops/hn-jobs-data - Site: https://github.com/gantryops/hn-jobs-site

AI agents can consume the data directly there's an llms.txt at https://hn-job-trends.gantryops.dev/llms.txt with all the JSON endpoints and schemas.