LevelUp
An intelligent recruitment ecosystem designed to bridge the gap between candidate qualifications and interview performance through semantic matching and real-time behavioral metrics.
The Problem
Standard Applicant Tracking Systems (ATS) often fail both the recruiter and the candidate by relying on rigid keyword matching, which ignores the semantic context of a professional's experience. Furthermore, while technical skills can be analyzed on paper, soft skills and behavioral delivery remain invisible until the interview—where candidates are often least prepared. LevelUp was engineered to solve both by unifying CV analysis with a live behavioral coaching environment.
Technical Architecture
Semantic Matching Engine
Utilizes Sentence-BERT (SBERT) to transform unstructured PDF text into high-dimensional vector embeddings. By applying Cosine Similarity, the system calculates a conceptual match score between the candidate's CV and the Target Job description, successfully identifying qualified applicants even when terminology differs from the job posting.
Computer Vision Pipeline
Integrates MediaPipe Face Landmarker directly in the browser to track behavioral indicators. The system monitors gaze direction (eye contact), facial landmark stability (posture), and expression intensity. These metrics are processed locally to maintain privacy and low latency before being synthesized into the final performance report.
Dynamic Interview Flow
Employs the Web Speech API for real-time speech-to-text conversion. The transcript is sent to a Gemini-powered backend that generates context-aware follow-up questions based on the candidate's specific resume and previous answers, creating a unique, non-linear interview experience.
Django REST Core
The backbone is a robust Django framework managing complex relational data between users, resumes, and interview sessions. It handles asynchronous AI report generation and serves as the secure vault for the platform's authentication and job-matching logic.
Automated Testing
Rigorous Cypress E2E suites ensure 100% success rates in the "handshake" between the React frontend and Django backend, especially during complex multipart form uploads and live video session initialization.
Real-time Sync
Implemented custom React hooks to manage the complex state transitions between the interview countdown, active recording, and the final data-heavy grading analysis phase.
Data Integrity
Ensures PDF layouts are normalized using specialized text extraction pipelines, allowing the AI to accurately identify skills even in non-standard resume formats.
The Result
LevelUp successfully demonstrates how high-level AI concepts—like semantic vector search and real-time computer vision—can be unified into a practical, user-centric tool. The platform doesn't just analyze; it coaches, providing a holistic 0–100 performance score that reflects both the technical strength of the CV and the soft-skill delivery of the interview.