Experience

Designed scalable ML training pipelines for the DentalScan platform, processing 5,000+ dental images with structured preprocessing, augmentation, and validation for disease classification and region-level detection. Developed evaluation utilities in PyTorch to compute classification and detection metrics including precision/recall, confusion matrices, and IoU-based scoring, reducing experiment comparison time by 30%. Integrated ML inference outputs with backend APIs via versioned response schemas and consistent payload formats, enabling automated result delivery. Implemented model versioning and deployment support with structured logging and config-based runs, improving reproducibility across retraining cycles.

Designed and maintained an ML training-data pipeline using AWS S3, EC2, and PostgreSQL, managing structured datasets and improving data retrieval efficiency by 35% for model training workflows. Developed backend APIs using Python (FastAPI) and containerized with Docker, reducing result-serving latency from 900ms to 280ms via query optimization and indexing. Integrated backend APIs with the frontend dashboard to display casino fairness rankings, reducing analysis time by 30%.

Contributed to feature enhancements and bug fixes for a React/MySQL academic portal used by 2,500+ users daily, improving reliability across attendance and staff workflows. Implemented attendance automation features (mark-all and quick-edit), saving approximately 8 hours per week per department across 5–6 departments. Optimized results-page performance by refactoring slow JOIN-heavy queries with MySQL indexing, reducing load time from 3.2s to 1.0s. Strengthened security for grade and results pages by implementing RBAC and OAuth2.0, preventing unauthorized access.