All projects
Felt cover for Seperable

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.

private code

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