BMATS

System Design Diagram

Inspiration

The recruitment process is often inefficient and prone to algorithmic bias. Our core assumption is that this bias is inherent in most current Applicant Tracking Systems (ATS), which primarily rely on literal keyword matching. These systems scan a job posting and a resume, filtering candidates based on exact keyword overlap. This rigid approach often fails to recognize synonymous terms or related concepts, leading to a critical flaw: qualified candidates are prematurely "silenced" and filtered out simply because their resumes are not "ATS-friendly."

This project was inspired by the universal challenge of ensuring fair consideration for all applicants. We aim to address this systemic issue by moving beyond keyword searching. By implementing a semantic analysis system, we can analyze the contextual meaning of a candidate's experience, identify skills that are synonymous with job requirements, and provide a more holistic and valid evaluation of their legitimacy for a position. Our goal is to bring previously silenced, qualified candidates back into focus and create a more equitable hiring landscape.

What it does

BMATS (Bias-Mitigating Applicant Tracking System) is designed to reduce noise and bias in the initial stages of recruitment. The system targets HR departments and hiring managers by providing a more equitable and data-driven approach to candidate evaluation.

It leverages Natural Language Processing (NLP) to anonymize personally identifiable information (PII) from resumes and applies a hybrid semantic ranking algorithm. This process highlights candidates based on skills and experience relevant to a job description, rather than on personal details. Therefore, it surfaces the talent that might otherwise be missed.

How we built it

Our system architecture comprises two main components: an NLP processing pipeline and a lightweight web application.

NLP Pipeline:
- PDF Parsing: Resumes are ingested and parsed using the PDFPlumber library.
- Anonymization: Personally Identifiable Information (PII) is detected and redacted using spaCy's Named Entity Recognition (NER) models.
- Semantic Ranking: The anonymized text is processed by a cross-encoding models, which scores candidate profiles against job requirements.
- Data Integration: The processed data and resulting scores are stored in a structured format within our database.
Lightweight Web App:
- Frontend: A responsive and intuitive user interface built with React.js.
- Backend: A robust API server developed with FastAPI, providing efficient endpoints for data handling and communication with the NLP pipeline.
- Database: A SQLite database for lightweight, serverless data storage and management.

Challenges we ran into

Several challenges were addressed during the development cycle:

Conceptualization: Defining a clear and impactful project scope during the ideation phase.
System Design: Architecting database schema and designing RESTful API routes.
Edge Case Management: Implementing comprehensive testing to identify and resolve edge cases, such as handling non-standard PDF layouts and uncommon PII formats.
Research and Development: Configuring an open-source ATS model that has algorithmic bias for testing purposes.

Accomplishments that we're proud of

Collaborative Environment: Successfully simulated a real-world development environment, emphasizing teamwork, communication, and working effectively under pressure.
UI/UX: Delivered a polished and user-friendly web prototype that exceeded initial expectations.
Technical Exploration: Gained valuable experience with new libraries and frameworks, particularly in the domains of NLP and web development.

What we learned

Agile Methodologies: The importance of task-driven development and iterative planning in achieving project milestones.
Team Synergy: Effective communication is critical to overcoming obstacles and ensuring alignment within a development team.
Technical Proficiency: Deepened our understanding of system design, NLP techniques, and full-stack development.

What's next for BMATS

Future development will focus on enhancing the platform's capabilities and robustness:

Secure Authentication: Implementing a comprehensive, production-ready authentication and authorization system.
Cloud Deployment: Migrating the application to a scalable cloud infrastructure (e.g., AWS, Azure, or GCP) to ensure high availability and performance.
Feature Expansion: Introducing new features