Skip to content

ARCHITECTURE.md

Latest commit

 

History

History
67 lines (49 loc) · 2.06 KB

ARCHITECTURE.md

File metadata and controls

67 lines (49 loc) · 2.06 KB
Raw

Architecture

JustHireMe is a local-first desktop app with a Tauri shell, React frontend, and Python backend sidecar.

High-Level Flow

Profile ingestion
  -> Kuzu graph + LanceDB vectors
  -> Source scrapers
  -> Lead quality gate
  -> Fit ranking / semantic matching
  -> Customization package generation
  -> Local CRM and review UI

Frontend

The React app in src/ is responsible for:

  • navigation and workspace UI
  • lead cards and filters
  • settings
  • profile and ingestion screens
  • customization package review
  • WebSocket event display

The frontend talks to the backend through authenticated local HTTP requests. Tauri provides the backend port and API token at runtime.

Backend

The Python backend in backend/ is responsible for:

  • FastAPI routes
  • source scraping
  • quality gating
  • ranking and evaluation
  • profile ingestion
  • vector search fallback behavior
  • PDF and outreach generation
  • local persistence

Important modules:

  • backend/automation/free_scout.py: direct/free source scraping
  • backend/automation/scout.py: broader source scraping orchestration
  • backend/discovery/quality_gate.py: pre-save quality checks
  • backend/ranking/scoring_engine.py: deterministic fit scoring
  • backend/ranking/semantic.py: LanceDB semantic matching and fallback behavior
  • backend/generation/service.py and backend/generation/generators/: resume, cover letter, outreach, and package generation
  • backend/data/repository.py: repository facade for local persistence
  • backend/data/sqlite/, backend/data/graph/, and backend/data/vector/: SQLite, Kuzu, and LanceDB access helpers

Storage

Local storage includes:

  • SQLite for leads, settings, events, generated asset metadata
  • Kuzu for profile graph data
  • LanceDB for profile vectors
  • local files for generated PDFs

These files should not be committed or uploaded in public issues.

Experimental Automation

Browser automation exists as a contributor lab and is intentionally separate from the core OSS promise. The supported workflow is scraper, ranker, vector matching, and customizer.