Resu-Dog 🐾

AI-Powered Resume Optimization & Career Intelligence Platform

Upload your resume. Paste the job description. Get back an ATS-optimized resume, a recruiter-grade shortlist assessment, a skills gap analysis with a closure roadmap, and a personalized cover letter — in seconds.

---

Product Vision

Most people lose jobs before a human ever reads their resume. ATS systems filter out 75%+ of applicants automatically. Resu-Dog fixes that — and goes further. We don't sugarcoat. We give candidates a brutally honest, actionable picture of where they stand: what's working, what's missing, and exactly how to fix it.

The ethical line we hold: We help users present what they have more powerfully. We never fabricate skills, credentials, or experience. Trust is the product.

---

Core Features

1. Resume Upload

• Supported formats: PDF, DOCX, TXT
• Drag-and-drop or file picker
• Max file size: 10MB

2. Compatibility Score (Before & After)

• Targeted mode: Keyword Coverage, Skills Alignment, Formatting Compliance
• General mode: Clarity, Action Language, Structure, Completeness
• Independent before/after scoring — never in the same AI call
• After score teased on free tier — visible on paid

3. Optimized Resume

• Every bullet follows: [Action Verb] + [Specific Achievement] + [Measurable Outcome]
• Weak openers (Responsible for, Helped, Assisted) explicitly banned in prompts
• Keywords from JD naturally incorporated
• ATS-clean formatting — no tables, columns, broken headers
• Downloadable as PDF (paid tier)

4. Shortlist Assessment

• Recruiter-perspective verdict: Low / Fair / Good / Strong probability of shortlisting
• Strengths (referencing specific resume content)
• Critical gaps vs the JD
• 3 actionable quick wins
• Red flags a recruiter would notice
• Assessment-driven refinement: quick wins + gaps are automatically applied back into the resume before final delivery

5. Skills Gap Analysis (targeted mode only)

• Skills Heatmap: pill grid showing every JD skill as 🟢 Matched / 🟡 Transferable / 🔴 Gap
• Transferable Bridges: for each amber skill, one honest sentence mapping what they do have
• Gap Closure Roadmap: for each red gap, a specific cert / side project / course recommendation
• Collapses to a summary banner: "12 matched · 3 transferable · 2 gaps"

6. Matching Cover Letter (targeted mode only)

• Personalized — not a template
• Connects specific candidate experience to specific role requirements
• Professional "Dear Hiring Manager," salutation + sign-off with candidate's name
• Download/copy (paid tier)

---

Roadmap

✅ Phase 1 — Core Optimizer (shipped)

• Upload CV (PDF / DOCX / TXT) + paste job description
• GPT-4o rewrites the resume for ATS compatibility
• Independent before/after compatibility scores
• Cover letter generated from the optimized resume
• PDF download
• Time-aware theme: auto light (6 AM–6 PM) / dark (6 PM–6 AM) based on local time, manual toggle persists preference
• Next.js Route Handler — no separate backend required

✅ Phase 2A — JD-Optional Mode (shipped)

• Targeted mode: Upload CV + paste JD → scored against the JD
• General mode: Upload CV only → scored on general quality (Clarity, Action Language, Structure, Completeness)
• Mode toggle (pill UI) on optimize page

✅ Phase 2B — CV Builder Wizard (shipped, /build)

• Paste job description → AI generates targeted questions specific to that role
• User fills multi-step structured form (questions differ per role — not generic)
• AI writes full tailored CV + scores + shortlist assessment
• AI Assist per question ("Help me write this" / "Rewrite with AI")
• Assessment-driven refinement applied before final delivery

✅ Phase 2C — Shortlist Assessment + Assessment-Driven Refinement (shipped)

• Recruiter-grade assessment panel after every optimize/build
• refineWithAssessment() applies quick wins + critical gaps back into the resume
• Merged changes_made list shows both optimize + refinement improvements

✅ Phase 2D — Skills Gap Analysis (shipped)

• Skills heatmap (🟢 Matched / 🟡 Transferable / 🔴 Gap) as a scannable pill grid
• Transferable bridges: one honest sentence mapping existing experience to each amber skill
• Gap closure roadmap: specific named certs/courses/projects for each red gap
• Collapsible GapAnalysis.tsx card below ShortlistAssessment
• Targeted mode only — hidden in general mode (no JD to compare against)

---

✅ Phase 3 — Public Profile Page + Shareable Links (shipped)

Flow:

1. User completes optimize or build → sees results as normal
2. "Create Profile Page →" button appears in ResultsPanel header
3. One GPT call parses the optimized resume into structured profile JSON
4. Stored in localStorage → user redirected to /p/preview
5. "Share Profile" button publishes to Upstash Redis → generates shareable URL like /p/a1b2c3d4e5
6. Public visitors see the full profile with dynamic OG meta tags, Hire Me CTA, Download CV, and "Built with resu-dog.com" footer
7. Shared profiles auto-expire after 90 days

Page layout (Udemy-inspired, two-column):

Hero: Avatar (initials) · Name · Title · Location · Open to Work badge · ATS score
Main (scrolls): What I bring · Experience timeline · Education · Skills pill grid
Sidebar (sticky): Stats · Hire Me CTA · Share/Copy Link · Download CV · LinkedIn · "Made with resu-dog"

Animations (Framer Motion):

• Timeline sections fade in on scroll (IntersectionObserver)
• Stats count up on sidebar viewport entry
• Skill pills stagger in
• Smooth hero entrance

Export As — same optimized content, different formats:

Format	What changes
ATS PDF (current)	Clean plain text, ATS-parsed — 10pt body, 18pt name, neutral grey dividers
ATS DOCX	Calibri font, 0.75in margins, industry-standard formatting
LinkedIn Profile	Splits into Headline, narrative About, Experience, Skills, recommendation request templates
WES / Immigration CV	Credential equivalency framing, Canadian format, Month Year date ranges, GPA/honours
UK / EU CV	2-page norm, UK date format, hobbies only if present (never fabricated), "references available"

Each format = one extra GPT call with different formatting rules. Architecture: POST /api/v1/export with format field.

---

🚧 Phase 3B — Time-Aware Theme (shipped)

• Auto theme: First-time visitors get light or dark based on local time (6 AM–6 PM = light, otherwise dark)
• Manual override persists: Once user toggles the theme, their preference is saved in localStorage and sticks across visits
• No geolocation needed: Uses new Date().getHours() which respects the browser's local timezone
• Implementation: TimeBasedDefault child component inside ThemeProvider checks localStorage.theme on mount — only applies time-based default if no manual preference exists

🚧 Phase 4 — Auth + Paywall

• Supabase Auth (email + Google OAuth)
• Save optimize/build results per user
• Stripe paywall — remove isPaid={true} hardcode, real subscription tier

� Phase 5 — Platform Expansion Modes

Mode	Target Market
North American Resume (ATS)	Core product — Phase 1 ✅
LinkedIn Profile Optimizer	1B+ users, massive TAM
WES / Credential Evaluation	Immigrants, international students
UK / AU / CA CV Format	International job seekers
ATS-specific (Workday, Greenhouse, Lever)	Power users, recruiters

Each mode is a separate feature/upsell tier. The architecture already supports a profile_mode field in the data model.

---

Monetization Strategy

Model: Product-Led Growth (freemium → subscription)

The user experiences the value before being asked to pay. They see their low score and the improved score — then hit the paywall exactly when the pain of not getting the result is highest.

Free Tier (The Hook)

Feature	Access
Upload resume + paste JD	✅ Always free
Compatibility score before	✅ Visible
Score after optimization	⚡ Teased / blurred
Optimized resume download	❌ Locked
Cover letter	❌ Locked
Keyword breakdown detail	⚡ Partial
Optimizations per month	1–2

Paid Tier (~$12/month — 3-day free trial)

Feature	Access
Full optimized resume (PDF + DOCX)	✅
Full cover letter	✅
Complete keyword breakdown	✅
Resume history & versions	✅
Unlimited optimizations	✅
Multiple job applications tracked	✅
Public profile page (Phase 2)	✅
Profile view analytics	✅
Platform modes (LinkedIn, WES, etc.)	✅

The Conversion Moment

User uploads resume → sees score: 42/100
        ↓
System optimizes   → shows score: 87/100
        ↓
"Download your optimized resume" → [Sign up / Upgrade]

They have already seen the improvement. The pain of not getting it drives conversion.

Subscription note: Job hunting is not a daily activity, but when someone is actively hunting they apply to 10–30 jobs. A monthly subscription during that window is natural. Subscription beats one-time payment for LTV.

---

Tech Stack

Layer	Technology	Notes
Frontend	Next.js 16 (App Router)	TypeScript, Tailwind CSS v4
Auth	Supabase (planned)	Email + Google OAuth — Phase 4
AI / LLM	OpenAI GPT-4o	JSON mode, called via Next.js Route Handlers
File Parsing	mammoth + pdf-parse	PDF, DOCX, TXT — server-side in Route Handlers
PDF Output	jsPDF	Client-side dynamic import
DOCX Output	docx + file-saver	Client-side generation
Profile Storage	Upstash Redis	Shareable profile links, 90-day TTL
Scroll Animations	Framer Motion	Profile page + landing page
Database	Supabase (planned)	Phase 4+
Payments	Stripe (planned)	Phase 4+
Hosting	Vercel	Auto-deploy from GitHub, dev → main

---

Architecture Overview

User Input
  ├── Resume (PDF / DOCX / TXT)
  └── Job Description (paste)
          │
          ▼
   Next.js Route Handlers
  ├── mammoth       → DOCX text extraction
  ├── pdf-parse     → PDF text extraction
  └── plain text    → TXT passthrough
          │
          ▼
   AI Processing Layer (OpenAI GPT-4o)          [parallel where possible]
  ├── Score BEFORE optimization
  ├── Rewrite resume (action verbs + JD keywords, no fabrication)
  ├── Score AFTER optimization
  ├── Generate cover letter
  ├── Shortlist assessment
  └── Assessment-driven refinement (quick wins + gaps applied back)
          │
          ▼
   Output Layer
  ├── Score card (before / after)
  ├── Optimized resume (download PDF + DOCX)     — paid
  ├── Cover letter (download / copy)              — paid
  ├── Shortlist assessment + gap analysis
  ├── Export As (LinkedIn / WES / UK)
  └── Shareable profile page (/p/[slug])
          │
          ▼
   Storage Layer
  ├── localStorage   → profile preview + resume text (owner)
  └── Upstash Redis  → published shareable profiles (90-day TTL)

AI Prompt Design Principles

• Scoring and rewriting are separate LLM calls (accuracy)
• Structured output via JSON mode (reliable reassembly)
• Rewrite prompt explicitly instructs: only rewrite existing content, never fabricate
• All three post-parse calls run in parallel via asyncio.gather (speed)

ATS Formatting Rules Enforced

• No tables, no columns, no text boxes
• No headers/footers (many ATS parsers skip these entirely)
• Clear section labels: EXPERIENCE, EDUCATION, SKILLS, CERTIFICATIONS
• Standard fonts, no embedded images in resume body

---

Security & Privacy

• Resumes contain PII — clearly stated in privacy policy: data is not stored or used for training
• OpenAI API by default does not train on API call data
• File size limits enforced server-side (10MB max)
• File type validation on both client and server
• Env vars for all secrets — never committed to repo

---

Project Structure

resu-dog/
├── frontend/                        Next.js 16 app
│   └── app/                         App Router
├── backend/
│   ├── main.py                      FastAPI entry point
│   ├── requirements.txt
│   └── app/
│       ├── api/routes/
│       │   ├── optimize.py          Full pipeline (parse → score → rewrite → cover letter)
│       │   ├── score.py             Score only
│       │   ├── cover_letter.py      Cover letter only
│       │   └── profile.py          Phase 2 — public profile page
│       ├── services/
│       │   ├── ai.py               All OpenAI calls
│       │   └── parser.py           File text extraction
│       ├── models/schemas.py        Pydantic models (profile_mode field included)
│       └── core/config.py           Settings / env config
└── .github/workflows/
    ├── ci.yml                       Lint + type check on all pushes
    ├── deploy-preview.yml           dev branch → Vercel preview URL
    └── deploy-production.yml        main branch → Vercel production

---

Branch Strategy

Branch	Purpose	Deploys To
dev	Active development	Vercel Preview URL
main	Stable production releases	Vercel Production

Workflow: all work on dev → PR to main → auto-deploy to production.

---

Local Development

Frontend

cd frontend
npm install
cp .env.local.example .env.local   # fill in NEXT_PUBLIC_API_URL
npm run dev                         # http://localhost:3000

Backend

cd backend
python -m venv venv
source venv/bin/activate            # Windows: venv\Scripts\activate
pip install -r requirements.txt
cp .env.example .env                # add OPENAI_API_KEY
uvicorn main:app --reload           # http://localhost:8000/docs

---

Deployment Setup Checklist

• Connect repo to Vercel, set frontend/ as root directory
• Add VERCEL_TOKEN to GitHub → Settings → Secrets → Actions
• Deploy backend to Railway or Render
• Add OPENAI_API_KEY to backend environment variables
• Add NEXT_PUBLIC_API_URL (backend URL) to Vercel environment variables
• Configure Supabase project + add DB connection string
• Set up Stripe account + add keys to both environments