Resume2Job – AI-Powered Resume & JD Matching Assistant (RAG-based)

Resume2Job is a full-stack AI tool that builds a personalized assistant works exclusively on your uploaded Resume and Job Description (JD). It uses a custom Retrieval-Augmented Generation (RAG) pipeline to deliver highly relevant, context-aware responses for job seekers and students giving insights that generic models or keyword-based ATS systems simply can’t.

✨ Key Features

Feature	Why It's Unique
🎯 RAG Personalized AI	Works only on uploaded resume + JD, not general internet data.
📄 JD Input Flexibility	Upload JD as file or paste as text (auto-saved for processing).
🧠 Semantic Matching	Uses vector embeddings (FAISS + metadata) to retrieve meaningful context.
🪄 ChatGPT-Like UX	Clean, modern UI with chat-like interaction, dynamic responses.
🧪 Multi-format Support	Handles PDF, DOCX, TXT, and image (OCR) inputs for both resume and JD.
🔑 Session Isolation	All files and context tied to a unique session_id per user session.

📈 Use Cases

• 📄 Resume tailoring for specific job roles
• 🧠 Gap analysis (missing skills or mismatched experience)
• ✍️ Suggesting resume improvements based on JD keywords
• 🗣️ Practice prompts: "How do I improve this resume for this job?"

🧩 Problem It Solves

Generic AI tools and ATS systems often fail to provide resume-specific insights because they operate on generalized data or keyword matching.

Resume2Job changes that by:

• Deeply analyzing your resume and a specific job description side-by-side
• Fetching the most relevant sections semantically using vector embeddings
• Providing feedback, skill-gap analysis, tailoring suggestions — all in real-time

🚀 Demo Screenshot

(Add screenshot of the home page / chat interface here)

🧠 How RAG Works Here

Upload Resume + JD → Extract + Clean + Sectionizing → Chunk → Embed → Store(FAISS) ↓ User prompt → Semantic Search(Top-k chunks) → Inject into LLM prompt → Get Response

📂 Semantic Pipeline Components

• Extractor: Parses PDF, DOCX, TXT, and image (OCR via PyMuPDF & Pillow)
• Cleaner: Removes non-informative content (garbage, footers, broken headers)
• Sectioner: Detects & organizes sections like Skills, Experience, etc.
• Chunker: Splits resume and JD into semantic chunks for better embeddings
• Embedding: Uses Sentence Transformers to convert chunks to vectors
• FAISS Store: Stores embeddings with metadata, organized by session
• Retriever (RAG): Top-K semantically matched chunks are pulled at runtime
• LLM Prompting: Chunks + prompt are combined and sent to Google’s Gemini/Generative AI API

🖥️ Frontend – React.js (Deployed ✅)

(Add screenshot of upload UI and chat interface here)

Key Features

• Clean, modern, dynamic UI
• Upload Resume (PDF, DOCX, IMG)
• Upload or paste JD (with toggle)
• Dynamic chat interface with expandable input and real-time responses
• Built with Vite + TailwindCSS for speed and responsiveness

⚙️ Backend – Flask API (In Progress 🚧)

Endpoints

POST /upload

• Accepts: session_id, resume, jd_file or jd_text
• Stores them under uploaded_files/<session_id>/

POST /prompt

• Accepts: session_id, prompt
• Retrieves stored resume + JD
• Passes through extraction → chunking → vector search → LLM

🧪 Tech Stack

🔨 Frontend

• React.js (Vite)
• Tailwind CSS
• react-dropzone for file uploads
• axios for API integration

🧠 Backend

• Flask 3.1 + Flask-CORS
• FAISS for vector similarity search
• Sentence Transformers (for semantic embeddings)
• PyMuPDF, Pillow, python-docx, python-magic (file parsing)
• Google Generative AI (for LLM inference)

📦 Python Dependencies

Flask==3.1.0
flask-cors==5.0.1
faiss-cpu==1.11.0
numpy==2.3.1
sentence-transformers==5.0.0
requests==2.32.4
python-dotenv==1.1.0
gunicorn==21.2.0
google-generativeai==0.8.5
tqdm==4.67.1
python-docx==1.2.0
python-magic==0.4.27
Pillow==11.1.0
PyMuPDF==1.26.1

🔐 Session Management

Each user interaction (uploads + chat) is assigned a unique session_id:

• Ensures file isolation
• Enables easy testing/debugging
• Scalable across users or environments

🧑‍💻 Author & Developer Notes

• Built completely from scratch, including:
• Custom file parsers for multiple formats
• Smart semantic chunking and embedding logic
• Scalable FAISS setup with dynamic metadata
• All pre-processing, chunking, and semantic retrieval were built from scratch no AI or LLM involvement until the final prompt stage.

🤝 Let's Connect

If you found this project helpful or are interested in collaborating, feel free to connect with me on LinkedIn.

📝 License

This project is licensed under the MIT License.