A modern AI-powered job matching system that bridges the gap between resumes and job descriptions using Natural Language Processing and deep learning. Built with a Two-Tower Embedding Model, this project enables accurate, scalable, and semantically rich candidate-job matching for the future of recruitment.

This project develops a job recommendation system that efficiently matches resumes to job descriptions using:

• Two-Tower Embedding Model
• Contrastive Learning
• Sentence-BERT (Transformer-based embeddings)
• TF-IDF keyword features
• FAISS for fast similarity search
• Cosine Similarity for match scoring

The system encodes job descriptions and resumes into a shared high-dimensional space, allowing for semantic similarity comparison beyond keyword matching.

1. Data Preprocessing

   • Clean and tokenize job descriptions and resumes
   • Apply Named Entity Recognition (NER) to extract key skills
   • Generate embeddings using Sentence-BERT and TF-IDF

2. Model Architecture

   • Two independent towers:
     • Job Tower: Encodes job descriptions
     • Resume Tower: Encodes resumes
   • Both towers map inputs into a shared vector space
   • Trained using contrastive loss to distinguish good matches from poor ones

3. Similarity Scoring

   • FAISS performs fast nearest-neighbor search in embedding space
   • Cosine similarity calculates match scores
   • Top resumes are ranked by semantic and skill-based relevance

4. User Query Support

   • Input: job description or personal resume text
   • Output: Ranked list of resumes (or jobs) with match percentage

✅ Deep learning-powered resume-job matching
✅ Semantic understanding using Sentence-BERT
✅ Keyword importance with TF-IDF
✅ Real-time search with FAISS
✅ Custom resume input supported
✅ Cosine similarity score for transparency

• Recruiters: Quickly find the best candidate for a job
• Job Seekers: See which job descriptions fit their resumes
• HR Tech Platforms: Build smarter job search engines and match systems
• Educational Institutions: Help students find suitable internships or job roles

• Fine-tuning transformer embeddings on domain-specific resume/job corpora
• Incorporating additional metadata such as years of experience, education level
• Exploring alternative loss functions (e.g., triplet loss, batch hard mining)
• Adding bias detection and fairness auditing
• Expanding support for multilingual job markets

This project was developed as part of the Natural Language Processing (1120-DS000-MSA-0234) module. Special thanks to all instructors and peers who provided valuable feedback.