← All work

Automation tool · 2026

Job Search Engine

A personal job-search engine that scans hundreds of company career pages and job boards every morning, scores each role with explainable rules and turns the good ones into a tracked application pipeline.

Demo with fictional data.

The situation

Finding the right roles meant checking dozens of career pages by hand every week, reading mostly irrelevant postings and losing track of applications.

What changed

One ranked inbox arrives every morning, with a plain-language reason for every score. Accepted roles become tracked applications with AI-prepared documents, and a human makes the final call.

Built for a senior professional running a focused job search across hundreds of companies.

Goals

  • Replace checking dozens of career pages by hand with one ranked inbox every morning.
  • Treat applications as a pipeline, from discovery and screening to preparation, application, interview and offer.
  • Make every score explainable, so it is clear why a role was suggested or filtered out.
  • Hand each accepted role to an AI-assisted preparation step, with a human always pressing submit.

What it does

  • Daily discovery across the major applicant tracking systems, venture-capital portfolio job boards and remote-job feeds, fetched in parallel. A failing source never stops the run.
  • Rule-based filtering by function, seniority, location policy, posting age and blocklists.
  • A 0 to 100 score with reasons written in plain language, covering bonuses for fit and penalties for red flags such as relocation or office attendance.
  • Deduplication across sources, past suggestions and existing applications.
  • An inbox where accepting a role creates a pre-filled application, and rejecting it records a reason and can blocklist the company for the future.
  • A tracker with a status rail, filters and a detail view per application with tabs for the job post, CV, cover letter, answers, company research, interview prep and timeline.

AI and automation

  • The engine runs on a schedule every morning, and catches up automatically if the computer was asleep.
  • For each accepted role, an AI agent drafts a tailored CV, a cover letter and form answers in the candidate's voice, then pre-fills the application in the browser for a final human review.
  • Generated documents are rendered to polished PDFs automatically.
  • Adding a new job board is pure configuration, with a request template and field map and no new code. Helper scripts probe and health-check every source.
  • The tool was built end to end with an AI coding agent.

Engineering notes

  • It uses the Python standard library only, with no frameworks, no dependencies and no build step, and JSON files serve as the database.
  • Writes are atomic, with automatic backups and locking, so the data survives crashes and concurrent updates.
  • It runs local-first on a loopback-only server, with strict rendering and file access rules.