Case study / Personal project
Job Hunter: automating the job-search pipeline
Job searching is an ETL problem wearing a motivation problem. Job Hunter automates the mechanical layer — aggregation, matching, tracking — so the human effort goes into the conversations that matter.
- Role
- Builder
- Timeline
- 2025
- Type
- Personal project
- Stack
- Next.js · MongoDB · LLM [VERIFY]
The idea
A serious job search means monitoring multiple boards, reading hundreds of near-identical listings, and remembering where every application stands. All of that is mechanical — which means it's automatable. Job Hunter ingests listings from multiple sources, deduplicates them, and scores each against a structured profile of skills, seniority and preferences using an LLM. [VERIFY]
How it works
- Ingestion — scheduled collectors pull listings from job boards and normalise them into one schema [VERIFY]
- Matching — an LLM scores each listing against the profile with reasons, not just a number, so the ranking is auditable
- Pipeline tracking — every application moves through explicit states (found → applied → interview → outcome) in MongoDB
- Digest — a ranked shortlist of genuinely new, genuinely relevant roles instead of an infinite scroll
Why it's on this site
It's a small system, but it's the same architecture thinking at personal scale: ingestion decoupled from scoring, scoring decoupled from presentation, and state made explicit. The patterns that keep a museum platform maintainable are the same ones that keep a weekend project alive.