Resume ⇔ Job Matching Engine

The Engineering Challenge: Recruitment processes were bottlenecked by manual keyword searching. The objective was to build an automated scoring engine capable of ingesting unstructured resume data and ranking candidates against job descriptions with high semantic accuracy.

Data Engineering & Feature Extraction:

• Automated Ingestion Pipeline:
  • Built a robust scraper using Selenium and BeautifulSoup to aggregate a training corpus of 15,000+ resumes and job descriptions.
  • Implemented a normalized storage schema in PostgreSQL (via SQLAlchemy), creating a reliable ground-truth dataset for reproducible training.
• NLP Feature Engineering:
  • Leveraged spaCy and NLTK for Named Entity Recognition (NER), extracting structured entities (Skills, Education, Location) from unstructured text.
  • Engineered custom interaction features, such as “Skill-Overlap Coefficient” and “Seniority-Gap,” combined with TF-IDF vectors to capture nuance beyond simple keyword matching.

Model Selection & Performance:

• Rigorous Benchmarking:
  • Evaluated multiple classifiers (Logistic Regression, Random Forest). LinearSVC proved superior for this high-dimensional sparse data, boosting the macro-F1 score by ~40% over the baseline.
  • Result: The final model achieved a 92% top-5 match accuracy across 20 distinct job categories, significantly reducing manual screening time.

MLOps & Production Deployment:

• Scalable Microservice Architecture:
  • Exposed the model via a high-performance FastAPI endpoint.
  • Generated automated Swagger/OpenAPI documentation to streamline integration for the frontend development team.
• Containerization & CI/CD:
  • Dockerized the application and implemented a GitHub Actions CI/CD pipeline pushing to AWS ECR.
  • Deployed on AWS ECS with configured auto-scaling policies based on CPU load to handle variable request traffic efficiently.

Technologies Leveraged: Python, scikit-learn (LinearSVC), spaCy, NLTK, FastAPI, PostgreSQL, Docker, AWS ECS, GitHub Actions.