📄 [PRD] Resume-Flow AI: Smart Job Application Assistant

Version: 2.1 (Comprehensive Technical & Product Draft)
Status: Ready for Development
Objective: To eliminate the friction of tailoring resumes and filling out repetitive forms for job seekers applying to 100+ positions, leveraging AI and automation without compromising user privacy or violating platform terms.

1. Executive Summary

This project is built on three core principles: Minimum Cost, Maximum Efficiency, and User Trust. Moving away from reckless “bulk apply” bots, we are building a semi-automated assistant that saves 90% of a job seeker’s time while strictly adhering to platform regulations.

Core Values:

• Zero-Storage Policy: Generated resume PDFs are stored exclusively in the user’s Google Drive, never on our servers. This ensures absolute privacy and reduces our cloud storage costs to zero.
• ATS-Proof Optimization (Core Metric): The ultimate goal is to bypass Applicant Tracking System (ATS) keyword filters. The AI cross-references the Job Description (JD) with the Master Profile, calculates an “ATS Match Score,” and visually demonstrates how targeted keyword injection increases compatibility (e.g., from a 45% to a 92% match) before the user applies.
• Surgical AI Integration: Instead of rewriting the entire resume (which risks hallucination and loss of user voice), the AI specifically injects missing JD keywords naturally into existing experience bullets or skill sections.
• Smart Assistant & Human-in-the-Loop: Beyond generating files, the Chrome Extension auto-fills application forms on job boards. However, the final “Submit” action and AI edits always require user confirmation (Semi-auto).

2. System Architecture (The 4 Pillars)

3. Core Functional Requirements

3.1 Authentication & Master Data Ingestion

• Clerk Auth: Users log in via Clerk across both the Extension and Web Portal. Clerk issues a JWT that securely communicates with the Backend and Supabase.
• Google OAuth: Acquired during onboarding to request the drive.file scope (creating/managing files only created by our app).
• Client-Side LinkedIn Parsing: To bypass LinkedIn’s strict anti-bot measures, the Chrome Extension reads the DOM of the user’s logged-in LinkedIn profile, extracts the text, and sends it to the backend to generate the “Master Profile.”

3.2 AI-Powered Resume Optimization & ATS Scoring (The Engine)

• Pre- and Post-ATS Match Scoring:
  • The backend analyzes the raw JD to extract required hard skills, soft skills, and certifications.
  • It calculates a baseline “ATS Compatibility Score” based on the user’s original Master Profile.
• Semantic Inference & Auto-Revise Logic: The LLM receives the Master Profile and JD. If a JD requires a specific skill (e.g., IIS) and the Master Profile lacks it but indicates related senior experience (e.g., 10 years as a .NET developer), the AI flags this for inclusion.
• Review UI (Anti-Hallucination): The extension displays a clear progress metric (e.g., a progress bar showing Before: 45% -> After: 92%) along with a quick diff of the text changes. The user can uncheck (reject) any inaccurate additions.

3.3 Client-Side PDF Generation & Sync

• On-the-fly Generation: Upon user approval of the AI edits, the Extension utilizes pdf-lib to map the final JSON data into a clean, pre-designed PDF template directly within the browser.
• Drive Sync: The extension (or backend via securely passed tokens) uploads the rendered PDF to the user’s Google Drive under Resume-Flow/[Company_Name]/ and retrieves a shareable link.

3.4 Smart Auto-fill (The Assistant)

• Form Detection: Content scripts detect standard input fields (Name, Email, Phone, Cover Letter, Work Experience) on platforms like LinkedIn Easy Apply or Workday.
• One-Click Fill: A floating “Auto-fill” button populates the inputs using the Master Profile and the newly AI-tailored summary.
• Fallback UX: If auto-fill fails due to unconventional HTML structures, the extension provides a sidebar with “Copy to Clipboard” buttons for every piece of data.

4. Data Model & Security Schema

4.1 Master Profile Table (master_profiles)

Hosted on Supabase. Strictly governed by RLS.

{
  "id": "uuid",
  "clerk_user_id": "string (e.g., user_2Qx...)",
  "full_name": "string",
  "email": "string",
  "base_skills": ["C#", ".NET", "SQL"],
  "experiences": [
    {
      "company": "string",
      "role": "string",
      "description": "string"
    }
  ],
  "created_at": "timestamp"
}

4.2 Application History Table (application_history)

Metadata only. No actual resume files are stored here.

{
  "id": "uuid",
  "clerk_user_id": "string",
  "company_name": "string",
  "job_title": "string",
  "drive_link": "string (URL)",
  "status": "enum (applied, interviewing, rejected)",
  "applied_at": "timestamp"
}

4.3 Database Security (Row Level Security - RLS)

• Supabase is configured to intercept the Clerk JWT.
• Policy logic: USING (requesting_clerk_user_id() = clerk_user_id).
• Result: A user (or any backend query made on their behalf) can only SELECT, INSERT, UPDATE, or DELETE rows where the clerk_user_id matches their own token. Cross-account data leakage is mathematically impossible at the DB level.

5. Detailed User Flow

1. Install & Onboard: User installs the Chrome Extension, logs in via Clerk, and connects their Google account.
2. Profile Ingestion: User visits their own LinkedIn profile; the Extension prompts to “Sync Profile.” Master data is saved to Supabase.
3. Discovery: User browses a job posting (JD) on LinkedIn or Indeed. The Extension detects the JD and activates the [Optimize for this Job] button.
4. The Optimization & ATS Scoring Loop:
   • Extension: Scrapes JD text -> Sends to Backend.
   • Backend: Extracts core requirements -> Calculates baseline ATS Score -> LLM injects missing keywords naturally -> Calculates new ATS Score -> Returns JSON.
   • Extension UI: Displays the ATS Match progress (e.g., 50% -> 90%) along with a quick diff of the text changes.
   • Action: User reviews the added keywords and clicks Approve & Generate.
   • Extension: Renders the ATS-friendly PDF -> Uploads to Google Drive -> Saves metadata to Supabase.
5. Application: User clicks “Apply” on the job board. The Extension’s Auto-fill button appears, populating all fields.
6. Tracking: User checks the Next.js Web Portal dashboard to view application history, statuses, and links to the Drive PDFs.

6. Non-Functional Requirements

• Privacy & Compliance (Right to be Forgotten): If a user deletes their account, a Clerk webhook triggers a cascading hard delete of all their data in Supabase. Drive files remain untouched (user’s property).
• Performance: The end-to-end process from clicking “Optimize” to the PDF being ready in Google Drive must take under 10 seconds.
• Cost-Efficiency: Utilize serverless functions (Vercel/NestJS) and generous free tiers (Supabase, Clerk) to maintain near $0 fixed infrastructure costs during the MVP phase.

7. Development Roadmap

• v1.0 (MVP): Single JD optimization from LinkedIn + ATS Match Scoring + PDF generation + Google Drive upload.
• v1.5 (The Assistant): Advanced form auto-fill support for major ATS platforms (Workday, Greenhouse, Lever).
• v2.0 (Batch Prep): “Cart” system to save multiple JDs and batch-generate optimized PDFs overnight.
• v3.0 (Premium): Subscription model integration, AI-driven interview prep based on the customized resume.

💡 Next Step

You can now share this Markdown file directly with teammates or AI coding agents such as Cursor or GitHub Copilot to provide complete project context.

The next helpful step would be either designing the Chrome extension UI/UX wireframes or writing the AI prompts and system prompt contract for implementation.