
Seperable
An application copilot that gathers internship and graduate listings, matches them to your profile, writes evidence-based CVs and tracks applications. Auto-submit is off by default.
Links
My role: Solo: everything, from scratch.
Problem
Applying for internships means reading hundreds of listings, rewriting the CV for each one and keeping track of replies. Mass-apply bots save time but send weak or untrue applications and break site rules.
Approach
A local, single-user pipeline collects listings, ranks them with Turkish full-text search plus vector search, and writes each CV only from approved facts in an evidence bank. Every automatic step sits behind gates: quotas, domain allowlists, verified form recipes, and a hard stop on CAPTCHA, consent boxes or sensitive fields. Auto-submit is off by default.
Result
Every stack choice was measured before it was kept, and the embedding test on 150 pairs replaced the first pick. The rule-based pre-filter drops 84% of listings before any LLM call, and a 30-listing hallucination test let no invented bullet reach a CV. The August 2026 report counts 1,147 unit tests.
Highlights
- 84% filtered before any LLM call
- 0 invented CV bullets in a 30-listing test
- 1,147 unit tests
- Auto-submit off by default
Stack
- TypeScript
- Next.js
- Tailwind
- BullMQ
- Redis
- PostgreSQL
- pgvector
- Drizzle
- Python
- Playwright
- Typst
- Gmail API
- Vitest