Self-hosted, single-user AI job-hunt workspace. Ingests job postings, scores them against your master CV with Claude or Gemini, tailors a per-role CV and cover letter, tracks applications and interviews, and aggregates the skill gaps that keep showing up across roles so you can close them.
Runs as two containers (Next.js + Postgres) bound to 127.0.0.1 on random, stable-after-setup ports.
- Job ingestion — paste a description, or live-search JSearch and LinkedIn (via RapidAPI) in one query.
- Match scoring — Claude or Gemini return a 0–100 score with strengths, gaps, and reasoning for every role.
- Tailored CV per role — rewrite prompt with canonical ATS section headings, warm summary, metric-driven bullets, skill-category grouping. Modern (colour + photo) or Classic (ATS-conservative) PDF theme, switchable per variant.
- Cover letters — per-application, template-driven, one-click generate-and-download PDF.
- Application tracker — sortable/filterable table (default) or Kanban board (toggle), with inline stage/priority/excitement/last-contact/notes edits.
- Interviews — multi-round scheduling with type (phone, video, onsite, technical, system design, behavioral, final), prep notes, calendar-ready fields.
- Interview flashcards — LLM-generated behavioral (STAR), role-specific, company-specific, and technical cards per job; flip-card UI.
- Feedback loop — structured post-interview log (what went well / badly / to change) with rejection-category taxonomy. Dashboard surfaces patterns after two entries.
- Cross-job gap tracker — aggregate skill gaps across every scored role, cluster synonyms via one LLM call, and generate curated learning links restricted to a whitelist of authoritative docs sites.
- LinkedIn SEO — score your headline and About against a target role; get keyword coverage, suggestions, and a ready-to-paste rewrite.
- LinkedIn PDF import — one-click import of your own LinkedIn profile as a new active master CV.
Match score, tailored CV with theme switcher, cover letter, flashcards, and interview scheduler — all on one page per job.
Every column inline-editable. Search + per-column filters. Toggle to a drag-and-drop board view.
One LLM call clusters every gap string from every match into canonical skills. The Learn drawer generates curated links from a whitelist of authoritative documentation domains — never any invented URLs.
Full walkthroughs and more screenshots in doc/starting.md.
git clone https://github.com/olafkfreund/rolehunter.git
cd rolehunter
./scripts/setup.sh # generates .env with random free ports and a DB password
$EDITOR .env # paste ANTHROPIC_API_KEY / GEMINI_API_KEY / JSEARCH_RAPIDAPI_KEY
docker compose up -d --build
docker compose exec app npx drizzle-kit migrate
source .env && echo "Open http://127.0.0.1:${APP_PORT}"To access from another machine, use an SSH tunnel — the app is intentionally bound to loopback.
doc/onboarding.md— what the app is, prerequisites, first-time setup flow.doc/starting.md— one section per page with every control explained, all 11 screenshots inline.doc/development.md— stack, directory layout, how to add columns / LLM methods / API routes / pages; coding conventions; screenshot regeneration.
Next.js 15 · React 19 · TypeScript · Tailwind v4 · Drizzle ORM · PostgreSQL 16 (pgvector) · Playwright (PDF) · Anthropic and Google Gen AI SDKs · TanStack Table · dnd-kit · sonner.
./scripts/backup.sh # writes backups/YYYYMMDD-HHMMSS/{rolehunter.sql.gz,uploads.tar.gz}To restore, pipe the SQL into docker compose exec -T db psql -U rolehunter rolehunter and extract the tar into the rolehunter_uploads volume.
Open issues describe every unimplemented feature and known infra gap. PRs welcome — keep them scoped to one issue where possible. See doc/development.md for conventions.