Job Engine Web

An AI-built job search pipeline, shipped end-to-end. Built solo with Claude Code as a working portfolio piece.

Live: job-engine-web.vercel.app Detailed spec: PRODUCT_SPEC.md — full feature map, API surface, schema, roadmap, change log (~1000 lines).

---

What it does

A personal mobile-first dashboard that consolidates a complete job-search workflow:

• Collect — pulls postings from Gmail alerts (LinkedIn, Seek) and the Adzuna API on a GitHub Actions cron (8 runs/day AEST). Atomic UPSERT into Supabase, dedup by hash.
• Rank — scores fit against my profile with Claude Haiku 4.5; returns a 0–100 score plus a reason.
• Tailor — generates a per-job resume diff with Claude Sonnet 4.6. Outputs [ORIGINAL] / [REVISED] pairs that match the actual base resume text in profile/*.md.
• Apply — copies the base Google Doc and runs the diff via the Docs API (batchUpdate.replaceAllText). Formatting preserved.
• Track — application status moves through Drafts → Submitted → Interview → Offer in Supabase, with a fire-and-forget sync to Google Sheets.

Architecture

Gmail / Adzuna  →  GitHub Actions  →  Supabase (seen_jobs)
                                        │
                          /api/rank (Haiku)
                                        │
                          /api/generate-resume (Sonnet)
                                        │
                          /api/docs/copy-base (Google Docs batchUpdate)
                                        │
                          applications table  ←→  Google Sheets

Frontend: Next.js 16 App Router on Vercel. Auth: Supabase SSR + Google OAuth, allow-list of one. Public landing page in front of the auth wall; all API routes return 401 to unauth'd callers via middleware.

Stack

Layer	Tech
Frontend	Next.js 16, TypeScript, Tailwind v4, PWA (iOS Add-to-Home-Screen)
Backend	Vercel Functions (App Router route handlers)
Database	Supabase (PostgreSQL) — seen_jobs, applications, collection_runs
AI	Anthropic Claude — Haiku 4.5 (ranking), Sonnet 4.6 (resume tailoring)
Storage	Google Drive (OAuth — file copy) + Google Docs API (Service Account — text replace) + Google Sheets (Service Account — pipeline sync)
Collection	GitHub Actions cron (Python scripts → Supabase)
Hosting	Vercel (Hobby)

How it's built

Spec-driven workflow with Claude Code. Every feature starts as a written spec in PRODUCT_SPEC.md; Claude Code authors the implementation under direction. I own the architecture, schema, and review.

Status

• Phase 5 ✅ Cloud collection cron + Profile tab + Search tab
• Phase 5.1 🟡 PWA install on iOS — push notifications deferred
• Phase 5.2 ✅ Top-picks lifecycle (active → applied → pipeline)
• Phase 5.3 ✅ Stale job handling (is_expired + 14-day age filter)
• Phase 6 ⏳ Auth wall + public landing for portfolio positioning

Running locally

npm install
npm run dev          # http://localhost:3000
npm run build        # production build

Required environment variables (see PRODUCT_SPEC.md § 2 for full schema): NEXT_PUBLIC_SUPABASE_URL, NEXT_PUBLIC_SUPABASE_ANON_KEY, SUPABASE_SERVICE_ROLE_KEY, ANTHROPIC_API_KEY, GDOC_BASE_PJT, GDOC_BASE_DA / DS / DE, plus Google service-account JSON and Gmail OAuth tokens.

Project layout

app/
  page.tsx              ← public landing (this is what recruiters see)
  dashboard/page.tsx    ← personal dashboard (auth-gated)
  login/                ← Google OAuth entry
  auth/callback         ← Supabase OAuth code exchange
  api/                  ← route handlers (queue, rank, generate-resume,
                          docs, applications, sheets, jobs)
  _landing/             ← landing-only components (Bento + animations)
lib/
  supabase/             ← SSR + browser + middleware clients
  anthropic.ts          ← Claude client
  google.ts             ← Drive/Docs/Sheets clients
  projects.ts           ← Google Docs project block manipulation
  profile-context.ts    ← SKILLS_MATRIX + PROJECTS_INVENTORY
middleware.ts           ← single-user email allow-list, route gating
profile/                ← base resume markdown (DA / DS / DE variants)
scripts/                ← collection scripts (Python)

Author

Gayoung Dan (Ina) — Master of Data Science, Monash University (2025). Melbourne, Australia.

gayoung.dan.data@gmail.com

---

For the full engineering detail — every endpoint, every schema column, every constraint, every phase decision — read PRODUCT_SPEC.md.