Spent 2.5 years building better job search (now using it to find a job)

Details

External ID

46647465

Source

HN

Company

—

Product

Spent 2.5 years building better job search (now using it to find a job)

Website domain

jsa.works

Launched

Jan. 16, 2026

Cohort

—

Upvotes

8

Upvotes percentile

0.41699604743083

Tags

—

Fetched at

Sept. 7, 2026, 9:25 p.m.

Updated at

Sept. 7, 2026, 9:25 p.m.

Description

Hello HN! After 2.5 years of development, I'm sharing Job Search Assistant (https://jsa.works) - even though I still don't feel quite ready. It's an open alpha of a job search platform that actually matches positions to your resume properly.The problem: Major platforms like LinkedIn, Indeed, etc. have terrible search despite huge resources. Search for "senior backend engineer" and you'll get frontend internships in the top results. They optimize for engagement, not relevance.My solution: JSA uses LLMs for resume/job parsing and semantic vector search for matching. Upload your resume, set filters, get jobs that actually fit your profile. Clean interface, no noise. Only fresh jobs (15-day retention) to avoid stale listings for now. Freemium model - essential search/filters are free with reasonable limits, paid tier adds kanban-style application tracking.Tech stack: ~78k lines of Go backend organized as microservices (scraper, indexer, searcher, etc.) communicating via NATS. Qdrant for vector search, PostgreSQL for relational data. HTMX frontend (shoutout to my friend @romshark who introduced me to HTMX - I'm not a frontend expert, so this is where AI agents helped to finish it). Scraping with go-rod. Self-hosted on a mini-PC in my utility room with scrapers running on Raspberry Pi - no cloud, just bare metal. Only SSO via Google/Microsoft for now.The scraping challenge: Modern job boards have sophisticated anti-bot measures. I built a simple deterministic fingerprint generator (https://github.com/chinese-room-solutions/fakebro) using Wave Function Collapse-like generation to create coherent browser profiles from a seed - matching user agents, Client Hints, and WebGL renderers that correspond to real hardware. The platform scraper rotates Chrome versions with unique fingerprints and handles Cloudflare challenges.Current status: Amsterdam and Paris only (data collection is expensive). If there's demand, I'll expand to EU and beyond. I'm using it myself right now to job hunt in those cities.Open alpha means bugs are expected, but I'm actively improving stability. The codebase is mostly pre-2025 human-written code, though AI agents helped me push through to completion in late 2025 after I went through the literal hell, mentally, and managed to stay alive and almost recover by the end the year.Fun story: Google suspended the project's GCP account citing "policy violations" with zero details or successful appeals, so I'm running Google SSO from my personal account for now. Classic cloud provider experience these days.Would love feedback, especially from folks in Amsterdam/Paris who could use this!

• Source listing
• Website