An AI-powered job application platform that helps candidates go from job posting to polished, tailored application documents. AutoApplicant crawls and imports job postings, maintains a structured career profile, and uses AI to generate tailored CVs, cover letters, and ATS (Applicant Tracking System) reports - exported as professionally rendered PDFs.

• Career profile management - structured profile with experience, education, skills, and projects; can be bootstrapped by parsing an existing CV
• Job discovery - multi-source job crawler (Jobindex, Jobnet, IT-Jobbank, Jobdanmark, + ATS boards) and Postgres full-text and pgvector semantic search
• LinkedIn job connector - personal-use, low-volume connector over LinkedIn's public jobs-guest endpoints, driven by LLM-generated per-user keyword plans; runs on its own jittered schedule off the shared crawl (see docs/guides/db.md / application.yml app.linkedin.*)
• AI document generation - tailored CVs and cover letters generated against a specific posting, with a configurable automatic drafter→reviewer loop that critiques and revises each draft before assembly
• ATS reports - automated analysis of how well a generated document matches the target posting
• Prompt-safety hardening - anti-fabrication rules (incl. tool-of-trade conflation), a prompt-injection guard treating scraped/posted job text as untrusted data, and a deterministic fact gate that flags invented/inflated metrics not supported by the profile (model-free, zero token cost)
• Pluggable AI providers - OpenAI, Gemini, or a local CLI-agent (Claude Code / Codex) for generation to run on a flat-fee subscription instead of API calls; embeddings always use a real API
• Structured document pipeline - all AI output is structured JSON (never raw text blobs), assembled server-side into a StructuredDocument with identity, sections, and rendering options
• PDF export - ATS-friendly and designed templates rendered server-side
• Privacy by design - personally identifying fields (name, email, phone, photo, links) are never sent to the AI provider; identity is merged into documents after the AI call
• Authentication - Firebase JWT; optional LinkedIn integration
• API documentation - full OpenAPI spec with Swagger UI

The backend strictly follows hexagonal architecture - every cross-boundary interaction goes through a port interface:

adapter/            Spring controllers, JPA adapters, AI client, crawler, PDF renderer
  web/controller/   REST endpoints - depend on port/in interfaces only
  persistence/      JPA entities + adapters implementing port/out
  ai/               OpenAI client implementing AiProviderPort
  crawler/          Job posting crawler
  pdf/              PDF rendering
port/
  in/               Use case interfaces (what the application can do)
  out/              Repository/external service interfaces (what the app needs)
usecase/            Business logic - implements port/in, depends only on port/out
domain/             Pure records/value objects - no framework dependencies

The frontend mirrors this discipline: core/api (HTTP services), core/models (interfaces mirroring backend records), features (routed components), shared/components (presentational).

Copy-Item .env.example .env   # then fill in secrets
docker compose --env-file .env -f infra/docker-compose.yml up --build

# Dependencies only
docker compose --env-file .env -f infra/docker-compose.yml up postgres -d

# Backend
./gradlew :backend:bootRun

# Frontend
cd frontend; npm install; npm run start:local

• OPENAI_API_KEY - OpenAI key with access to gpt-4o and text-embedding-3-small
• Firebase service account JSON at .secrets/firebase-service-account.json

Optional: DB_*, LINKEDIN_CLIENT_ID/SECRET, ALLOWED_ORIGINS (all have local defaults).

AI provider / feature toggles (all optional, sensible defaults):

• GENERATION_AI_PROVIDER - openai (default) · gemini · claude-cli / codex / cli (local agent). ENRICHMENT_AI_PROVIDER must stay a real API (produces embeddings).
• AI_CLI_COMMAND - CLI invoked for generation when using a local agent (default claude -p; prompt piped to stdin).
• AUTO_REVIEW_ENABLED - automatic reviewer critique/revise pass after generation (default true; each pass is one extra LLM call).
• FACT_GUARD_ENABLED / FACT_GUARD_MODE - deterministic fact gate on generated metrics (warn default, or block).
• LINKEDIN_SCRAPER_ENABLED, LINKEDIN_LOCATIONS - LinkedIn job connector (see app.linkedin.* in application.yml).

Full docs live in docs/ (see the index): specs/, architecture/, guides/ (setup, commands, testing, db), product/ (strategy, features, the Danish-market playbook), and archive/ for superseded material. Agent guidance is in CLAUDE.md.

./gradlew :backend:test          # backend
cd frontend && npm test          # frontend
cd frontend && npm run lint      # lint