ResumeForge

ResumeForge is a job-search assistant that uses retrieval-augmented matching to find your best-fit resume for a given job description, then helps you customize it and generate a cover letter with an LLM.

How it works

1. You upload one or more versions of your resume (PDF, DOCX, or TXT).
2. Each resume is parsed to text, embedded with sentence-transformers (all-MiniLM-L6-v2), and:
   • the vector is stored in Pinecone,
   • the extracted text and metadata are stored in PostgreSQL (hosted on Supabase),
   • the original file is stored in a Supabase Storage bucket.
3. When you pick a job description, ResumeForge embeds it the same way and queries Pinecone for the resume with the highest cosine similarity, scoped to your account.
4. From there you can use the built-in LLM integration (Groq) to tailor the matched resume to the job description and generate a cover letter.

Architecture

Frontend/   React 19 + TypeScript + Vite + Tailwind CSS
Backend/    Flask REST API

The frontend talks to the backend exclusively over VITE_SERVER_URL (see Environment variables). There is no server-side rendering — the Flask app is a pure JSON API (plus the Google OAuth redirect flow).

Backend layout

Backend/
  app.py                 # Flask app: routes only, delegates to Services/
  DB.py                  # SQLAlchemy engine/session + Supabase client
  CreateTable.py         # One-time script to (re)create DB tables
  Models/Models.py        # SQLAlchemy models: User, Document
  Services/
    UserService.py        # Auth, profile, password, account deletion
    DocumentService.py    # Resume upload, parsing, embeddings, grading
    PineconeService.py    # Vector CRUD + cosine-similarity search
    JobService.py         # Job search (JSearch API) + job detail lookup
  Prompts/
    LLM.py                # Groq client wrapper
    ResumeReview.py        # Improvement-suggestion prompt
    customResume.py        # Resume customization / cover-letter prompt

Frontend layout

Frontend/src/
  Pages/          One component per route (Login, Signup, JobList, JobDetails,
                  ResumeList, NewResumeEntry, Feedback, Profile)
  Components/     Shared UI (Navbar, Spinner)
  AppRouter.tsx   Route table

Features

• Email/password signup & login, plus Google OAuth
• Profile management: view/edit profile, change password, delete account
• Upload multiple resume versions per user
• RAG-based resume ranking against a job description (Pinecone cosine similarity)
• Job search via the JSearch API
• LLM-powered resume customization and cover letter generation (Groq)

Getting started

Prerequisites

• Python 3.11+
• Node.js 18+
• Accounts/keys for: Pinecone, Supabase (Postgres + Storage), Google Cloud OAuth client, Groq, RapidAPI (JSearch)

Backend setup

cd Backend
python -m venv .venv
.venv\Scripts\activate        # Windows
# source .venv/bin/activate   # macOS/Linux
pip install -r requirements.txt

Create Backend/.env with the variables listed below, then create the database tables once:

python CreateTable.py

⚠️ CreateTable.py drops all existing tables before recreating them. Only run it on first setup, or you will lose data.

Start the dev server:

flask --app app run --port 5000

Frontend setup

cd Frontend
npm install

Create Frontend/.env:

VITE_SERVER_URL="http://localhost:5000"

Start the dev server:

npm run dev

The app will be available at http://localhost:5173.

Environment variables

Backend/.env

Variable	Purpose
PINECONE_API_KEY, PINECONE_HOST, PINECONE_INDEX_NAME	Pinecone vector index
POSTGRES_URL / user, password, host, port, dbname	PostgreSQL connection (Supabase-hosted)
SUPABASE_URL, SUPABASE_KEY	Supabase client (Storage + Postgres)
OAUTH_CLIENT_ID, OAUTH_SECRET	Google OAuth client credentials
FLASK_SECRET_KEY	Signs the session cookie — required for login/signup to work
FRONTEND_ORIGIN	Frontend URL, used for CORS and post-OAuth redirects (default http://localhost:5173)
IS_PRODUCTION	Set to true once deployed behind HTTPS with the frontend on a different domain than the backend. Switches the session cookie to SameSite=None; Secure so login persists across the cross-site requests a split frontend/backend deployment makes — without this, login will appear to succeed but the session won't stick. Leave unset (false) for local dev.
GROQ_API_KEY	Groq LLM API access
RAPIDAPI_KEY	JSearch API (job search)

Frontend/.env

Variable	Purpose
VITE_SERVER_URL	Base URL of the backend API

Google OAuth setup

In Google Cloud Console → APIs & Services → Credentials, add an Authorized redirect URI matching your backend host, e.g.:

http://localhost:5000/login/google/authorized

Known limitations

• The resume grading/feedback flow (/api/grade, the Feedback page) is not fully wired up yet.
• Deleting a user does not cascade-delete their resumes, Pinecone vectors, or Supabase-stored files.
• No automated test suite yet.