🧭 Executive Summary

NEXTGENRESUME is an AI-powered resume classification engine that matches candidate profiles to job descriptions using natural language understanding. This project leverages DistilBERT embeddings and cosine similarity to automate early-stage recruitment, drastically reducing screening time while improving relevance.

---

🔍 Current State & Gap Identification

Manual resume screening is time-consuming, subjective, and inconsistent. Traditional keyword-based ATS (Applicant Tracking Systems) often fail to understand semantic context and overlook high-potential candidates due to rigid filters. There's a clear need for:

• Context-aware matching
• Scalability in bulk resume processing
• Reduction in recruiter workload

---

📌 Methodology

Solution Approach

1. Extract resume content from PDFs using pdfplumber
2. Preprocess text to isolate skills, education, and project details
3. Generate embeddings for resumes and job descriptions via DistilBERT
4. Compute cosine similarity scores to rank relevance

Design Decisions

• Chose DistilBERT over BERT for faster inference with minimal accuracy trade-off
• Used cosine similarity for lightweight, effective matching
• CSV-based skill mapping for flexibility and extensibility

---

🧪 Evaluation Framework

Performance Metrics

Comparative Analysis

---

📊 Dataset Sources & Description

• Resumes: Scraped, anonymized academic and industry resumes
• Job Descriptions: Curated from LinkedIn, Glassdoor, and job portals

Dataset Summary

---

🧹 Data Processing Pipeline

import pdfplumber
import re

with pdfplumber.open("resume.pdf") as pdf:
    text = ''.join([page.extract_text() for page in pdf.pages])

• Skills: Matched using regex and a predefined skill dictionary
• Education: Extracted using patterns (e.g., B.Tech, M.Sc., MBA)
• Cleaned & stored in CSV for traceability

---

🛠 Implementation Stack

---

🚀 Deployment Considerations

• Suitable for deployment via Flask/FastAPI backend
• Model can be containerized using Docker
• Supports batch processing or RESTful resume uploads

Monitoring & Maintenance

• Track similarity trends via logging
• Periodically retrain skill dictionaries and embeddings
• Can integrate with recruiter feedback for active learning

---

📈 Key Results & Interpretations

• High correlation between top resumes and job fit ratings (from HR test set)
• Recommender performance was consistent across multiple domains
• System favored resumes with strong project and skills alignment

---

⚠️ Limitations

• Does not extract work experience duration (future NER-based solution planned)
• Struggles with heavily styled resumes or scanned documents
• Domain-specific job roles (e.g., healthcare, legal) need fine-tuned models

---

🧠 Insights & Implications

• Semantic search in HRTech is achievable at scale
• Lightweight transformer models like DistilBERT offer real-time usability
• This model architecture is adaptable for LinkedIn Job Matching, Gig Platforms, or University Placements

---

🏢 Industry Relevance

This solution addresses a core recruitment pain point in:

• Startups and SMEs without dedicated HR teams
• Campus Placement Cells
• Online Job Portals seeking AI-based shortlisting

📚 Source Credibility

• Model: DistilBERT on Hugging Face
• Dataset: Scraped via job portals and anonymized academic CVs
• Inspiration: CV-JD Matching Repo

---

🌟 Uncommon Insights

• Resumes with project narratives outperform generic ones
• Cosine similarity boosts effectiveness when embedding length is normalized

---

📌 Visual Architecture

[Resume PDFs] + [Job Descriptions]
         ↓
 [Text Extraction → Cleaning]
         ↓
 [DistilBERT Embeddings]
         ↓
 [Cosine Similarity Matrix]
         ↓
 [Top-N Recommendations]

---

🧾 Summary of Findings

• Resume-job alignment benefits greatly from context-aware embeddings
• Simple similarity scores outperform traditional keyword ATS
• This project showcases a plug-and-play model for intelligent resume screening

---

💥 Conclusion: Why It Matters

This project is a practical, low-cost implementation of AI in HR, transforming how organizations approach talent acquisition. It's scalable, explainable, and adaptable for real-world deployment—marking a strong entry into semantic recruitment solutions.

---

🔗 Repository

👉 GitHub: NEXTGENRESUME by Bhavya Srujana