Building an on-premise Anti-Fraud system powered by LLMs with end-to-end responsibilities

Achieved high-throughput document processing (1,000+ docs/hour) with parallel workers, S3 storage integration, and automated error handling

Developed multi-modal AI pipelines leveraging computer vision and LLMs for automated classification, extraction, and verification of 15+ medical document types

Built multi-stage automatic validation and verification systems ensuring 99%+ data integrity and regulatory compliance, reducing manual review by 80%

Engineered modular, scalable frontend architecture using Next.js, enabling seamless integration with backend services

Developed RESTful APIs using FastAPI with SQL Server integration, ensuring secure communication between frontend and backend systems

Designed semantic segmentation framework using multi-modal data for cultural heritage preservation

Developed pixel-wise segmentation models to accurately segment Borobudur temple reliefs using UNet, SegNet, PSPNet, DeepLabV3+, and YOLO

Established comprehensive evaluation protocol incorporating segmentation quality and reconstruction metrics

Led development of Android-based real-time driver monitoring application using Single Shot Detector (SSD) MobileNetV2

Implemented driver distraction and fatigue detection system with deep learning and image processing

Collaborated with cross-functional team of ML engineers, cloud computing engineers, and mobile developers

Enhanced technical expertise in deep learning, computer vision, and team leadership