belumkerja.com — Indonesian LinkedIn-style Jobs Board

Domain: belumkerja.com (literally: "not yet working") Stack: 100% Cloudflare (Worker + D1 + R2 + KV + Vectorize + Workers AI + Queues + Cron + Durable Objects) Code home: /home/ucok/web/belumkerja.com/worker-cf/ Sister project: beasiswa.net (siswa profile inheritance)

1. Positioning & Why It Wins

The wedge: the user just got a beasiswa.net scholarship → graduates → already has a verified profile → one tap to flip to "siap kerja" status. Zero re-onboarding. That's the entire moat.

2. Three User Roles

                    ┌────────────────────┐
                    │      USER (D1)     │
                    │  user_type column  │
                    └──────────┬─────────┘
            ┌──────────────────┼──────────────────┐
            ▼                  ▼                  ▼
    ┌──────────────┐  ┌───────────────┐  ┌─────────────────┐
    │   pencari    │  │   perekrut    │  │    pemberi      │
    │   (siswa)    │  │  (headhunter) │  │   (employer)    │
    └──────────────┘  └───────────────┘  └─────────────────┘
    Job seeker        Middleman/agent     Company / individual
    Inherits siswa    Multi-company       hiring direct
    profile from      pipeline mgmt
    beasiswa.net      Commission tracking

Single users table, user_type ENUM('pencari','perekrut','pemberi'), separate role-extension tables. One person can hold multiple roles (a senior dev moonlighting as a recruiter, etc.) — modelled as user_roles(user_id, role) join, not exclusive ENUM.

3. Architecture — 100% Cloudflare

Cross-site auth: beasiswa.net and belumkerja.com share the same phone-based OTP issuer and a small shared_users D1 (or KV mirror) keyed by phone hash. A siswa logging into belumkerja.com with their beasiswa phone gets their profile auto-imported on first login.

4. Database Schema (D1)

Inherited from beasiswa.net (replicated, not joined cross-DB)

• users — slug, phone, phone_hash, email, name, user_type, bio, is_verified
• student_profiles — education_level, school, university, faculty, major, GPA, graduation_year, province/city, achievements, skills, interests
• portfolio_items — title, description, item_type, url

New tables for jobs board

-- Extended profile (LinkedIn-like)
job_seeker_profiles(user_id PK, headline, summary, current_status, expected_salary_min, expected_salary_max,
                    work_preferences JSON, available_from, cv_r2_key, video_intro_r2_key)

experiences(id, user_id, company_id, role_title, employment_type, start_date, end_date,
            is_current, description, location, location_type)

skills(id, slug, name, category)
user_skills(user_id, skill_id, level, years, endorsement_count)
endorsements(id, from_user_id, to_user_id, skill_id, created_at)

certifications(id, user_id, name, issuer, issued_at, expires_at, credential_url, file_r2_key)

languages(id, user_id, language_code, proficiency)

-- Social / network
connections(follower_id, following_id, status, created_at)  -- mutual when status='accepted'
posts(id, user_id, body, media_r2_keys JSON, post_type, created_at)
post_reactions(post_id, user_id, reaction)
comments(id, post_id, user_id, body, parent_comment_id)
recommendations(id, from_user_id, to_user_id, body, role_title, status)

-- Companies (employers)
companies(id, slug, name, industry_id, size, founded_year, hq_city, website, logo_r2_key,
          verified, owner_user_id, claim_status)
company_followers(company_id, user_id)
company_reviews(id, company_id, reviewer_user_id, salary, rating, pros, cons, anonymous)

-- Jobs
job_categories(id, slug, name, parent_id)
industries(id, slug, name)
jobs(id, slug, company_id, posted_by_user_id, posted_by_role,  -- 'pemberi' or 'perekrut'
     title, category_id, employment_type, location_city_id, location_type,  -- onsite/hybrid/remote
     salary_min, salary_max, currency, experience_level, education_required,
     description_md, requirements_md, benefits_md, application_url, application_email,
     status, posted_at, expires_at, view_count, applicant_count,
     embedding_id  -- Vectorize key
)
job_skills(job_id, skill_id, required INTEGER)
applications(id, job_id, applicant_user_id, headhunter_user_id NULL, cv_r2_key, cover_letter,
             status, applied_at, viewed_at, response_at)
saved_jobs(user_id, job_id, saved_at)
job_alerts(id, user_id, query JSON, frequency, last_sent_at)

-- Headhunter pipeline
talent_pools(id, headhunter_user_id, name, description)
talent_pool_members(pool_id, user_id, notes, status)
placements(id, headhunter_user_id, candidate_user_id, job_id, status,
           commission_amount, commission_paid, placed_at)

-- Programmatic SEO
seo_pages(id, slug, page_type, params JSON, title, description, h1, body_md,
          last_built_at, last_visited_at, view_count, status)  -- pSEO output
seo_keywords(id, query, category, source, search_volume_estimate, harvested_at)  -- from Google Autocomplete
seo_articles(id, slug, keyword_id, title, body_md, hero_r2_key, author_user_id,
             published_at, view_count, ai_model, edited_by_human)
qna_questions(id, slug, question, category, source_keyword_id, view_count, answer_count)
qna_answers(id, question_id, body_md, author_user_id, ai_generated, upvotes, accepted)

-- Cities / geo (reuse beasiswa)
cities(id, slug, name, province, lat, lng, population)

-- Notifications & messaging
conversations(id, type, participants_hash)
messages(id, conversation_id, sender_user_id, body, media_r2_keys, read_at)
notifications(id, user_id, type, payload JSON, read_at)

5. LinkedIn-style Profile Surface

URL: /u/{slug} (consistent with beasiswa.net /profil/{slug} — different path, same canonical user)

Sections in order:

1. Hero — photo, name, headline, location, current status badge ("Sedang Mencari", "Bekerja", "Freelance"), connect/message CTA
2. About (summary)
3. Pengalaman (experiences)
4. Pendidikan (student_profiles + custom educations)
5. Keahlian & Endorsement (skills, endorsements)
6. Sertifikasi
7. Portofolio (inherited from portfolio_items)
8. Bahasa
9. Rekomendasi
10. Aktivitas / Postingan (last 5 posts)

All sections SSR'd into static HTML (CLAUDE.md rule: never empty divs filled by JS). Edit-in-place via inline forms with <dialog> modals + HTMX-style partial updates.

6. Job Board Features

For pencari (job seeker)

• Bottom nav: Beranda / Lowongan / Jaringan / Pesan / Profil
• Swipe-cards UI for jobs (Tinder-for-jobs) with "Lamar / Lewati / Simpan"
• 1-tap apply with auto-CV (PDF generated from profile, stored in R2)
• AI cover letter generator (Llama 3.3 70B)
• WhatsApp job alerts (daily/weekly digest)
• "Cocok untukmu" semantic match (Vectorize belumkerja-jobs × user's embedding)
• Application tracker with status timeline

For perekrut (headhunter)

• Talent search with filters (skill, location, GPA, university, salary expectation)
• Boolean + semantic search (Vectorize belumkerja-candidates)
• Talent pools (saved candidate lists)
• Bulk message via WhatsApp templates
• Placement tracker with commission ledger
• "Submit candidate to my client" workflow (creates applications row with headhunter_user_id set)

For pemberi (employer)

• Company page (claim flow if pre-existing scraped page)
• Post job with AI-assisted draft (Llama suggests title, salary range, skills based on description)
• Applicant inbox with AI ranking ("Top 10 cocok")
• Video screening questions (R2-stored responses)
• Verified-employer badge (CF Access + KYB doc upload)

7. Programmatic SEO — URL Patterns

Target: ~500K indexable pages within 6 months. Each must be SSR'd, semantic, with FAQ schema and breadcrumbs.

Total indexable: ~310K core pages + ~80K Q&A + ~10K articles = ~400K. Once content engine runs steady-state, it'll cross 500K within 6 months.

Content quality guard

Empty /lowongan/{kota}/{kategori} pages with 0 jobs MUST NOT return 200 with thin content. Either:

• Show "0 lowongan saat ini, tapi 12 perusahaan di {kategori} aktif merekrut" + nearby city fallback + relevant articles, OR
• Return 404 if even that fallback yields nothing (Google penalizes thin doorway pages).

8. Article + Q&A Engine — Google Autocomplete

Pattern lifted from /home/ucok/web/ourfuntech.com/src/cron.ts:

// Hourly cron — harvest autocomplete, dedup, queue article gen
const seeds = [
  // city-specific
  ...cities.map(c => `lowongan kerja ${c.name}`),
  ...cities.map(c => `gaji di ${c.name}`),
  // role-specific
  ...roles.map(r => `cara menjadi ${r.name}`),
  ...roles.map(r => `gaji ${r.name}`),
  ...roles.map(r => `syarat ${r.name}`),
  // intent-specific
  'lowongan kerja fresh graduate',
  'lowongan kerja tanpa pengalaman',
  'tips wawancara kerja',
  'cara membuat cv',
  'cara melamar kerja online',
  // company-specific (top 200 companies)
  ...topCompanies.map(c => `cara melamar ${c.name}`),
];

for (const seed of seeds) {
  const r = await fetch(`https://suggestqueries.google.com/complete/search?client=firefox&q=${encodeURIComponent(seed)}`);
  const [, suggestions] = await r.json();
  for (const s of suggestions) {
    await db.prepare('INSERT OR IGNORE INTO seo_keywords (query, category, source) VALUES (?, ?, ?)')
      .bind(s, classify(s), 'google_autocomplete').run();
  }
  await sleep(500);  // rate-limit ourselves
}

Then a second cron picks N unprocessed keywords, classifies each as article|qna|salary_page|career_guide, and dispatches to Llama 3.3 70B with role-specific prompts. Output is human-edit-flag-able (seo_articles.edited_by_human); we batch-publish only after a daily quality check (length > 800 words, contains H2s, has internal links).

Q&A engine specifics

Long-tail queries that read as questions ("apakah", "bagaimana", "berapa", "kapan", "di mana") become qna_questions. Each Q&A page:

• One AI-generated authoritative answer at top (with Article + FAQPage JSON-LD)
• Open thread for human answers (boosts dwell time + freshness signal)
• "Pertanyaan terkait" cluster (3-5 from same category)

This is identical to the existing pattern on anehtapinyata.com and alamgaib.com — just retargeted to job intent.

9. Mobile-Native UX (per CLAUDE.md universal standards)

• Bottom nav 5 items, sticky, safe-area-inset
• NO hamburger menu
• Pull-to-refresh on /lowongan and /beranda
• Skeleton loaders, not spinners
• Horizontal carousels for "Lowongan untukmu", "Perusahaan trending", "Lulusan terbaru"
• Swipeable tabs on job-detail page: Deskripsi | Perusahaan | Pelamar | Q&A
• View Transitions API for navigation
• PWA: manifest, install prompt after 2nd visit
• Web Push: "lowongan baru cocok buatmu", deadline reminders
• Voice search on /lowongan (SpeechRecognition API)
• AI chat ("Tanya Belum") floating button — answers job/career questions, contextual to current page

10. Phased Rollout

11. Open Decisions (need your input)

These three shape the product more than anything else:

A. Profile sync model with beasiswa.net — LOCKED: Lazy-on-graduation

When a phone hash logs in to belumkerja.com for the first time, the worker reads beasiswa-db (second D1 binding, read-only intent) by phone_hash. If student_profiles.graduation_year <= currentYear, we one-tap import (copy users + student_profiles + portfolio_items rows into belumkerja's D1). After import, profiles diverge — beasiswa edits don't auto-flow into belumkerja. User can re-trigger import manually if they want fresh data.

Why: simplest implementation (no Queue, no schema coupling), respects the "graduation = transition" semantic, and avoids polluting belumkerja with still-studying users who aren't ready to job-hunt.

Edge case: phone exists in beasiswa but graduation_year > currentYear → show "Kamu masih kuliah. Datang lagi setelah lulus 🎓" with a "Notify me when ready" button (writes to a pre_register KV that fires a cron-driven WhatsApp on grad date).

B. Monetization

1. Free everything, ads only. Lowest friction.
2. Free for seekers, paid job posts for employers (Rp 200K-500K per post). Standard.
3. Free for seekers + employers, paid for headhunters (commission % on placement, or subscription for talent search). Aligns with the headhunter being the only party with strong ROI.
4. Freemium employer (3 free posts/month, paid for premium placement + AI ranking).

C. Headhunter business model

1. Open marketplace — any registered headhunter can claim/contact any candidate. Volume play.
2. Curated — headhunters apply + are vetted; quality > quantity.
3. Verticalized — only certain roles (e.g., tech, healthcare) initially; expand later.
4. Commission split — platform takes X% of every placement; headhunters can't operate "off-platform" with leads they got here.

Pick one per row (or write your own variant) and I'll wire the schema + features around your choices in the next iteration.

⚙ HARD CONSTRAINTS (enforced for all sites)

This domain MUST operate within these constraints — no exceptions:

• 100% Cloudflare serverless — Workers + D1 + R2 + KV + Workers AI + Vectorize. NEVER PM2, NEVER VPS, NEVER Docker in production path.
• 100% AI-automated — every customer interaction, every moderation decision, every transaction reconcile = AI. No manual queue, no live human chat support, no physical fulfillment.
• 1-operator solo — one person can run the entire operation from a phone. No team meetings, no shared inbox, no shift rotation.
• WhatsApp AI bot for all support (24/7, instant response, no SLA promises that need humans).
• Mayar QRIS for all Indonesian payments (subscription auto-renew, no manual invoicing).
• Indonesian UI primary — bahasa-first, English fallback only where unavoidable.
• Privacy — opt-in only, delete-on-request honored within 24h (cron-driven).
• No physical goods, no inventory — digital products + affiliate referrals only.

If the plan above describes any flow that violates these constraints, treat the plan as ASPIRATIONAL only and rework before building. The constraint trifecta wins.