sanaullah ashfat

Led the Data Analytics team at MetLife Bangladesh, designing a predictive algorithm for employee retention with an 82% F1 score. Collaborated with MetLife India to implement personalized recommendations in the 360Health app using user behavior analytics.

Engineered a Bangla Speech-to-Text system using OpenAI’s Whisper model, fine-tuned for dialectal variations to enhance recognition accuracy. Designed a real-time, scalable speech recognition pipeline with phoneme-aware pre-processing and robust error correction.

Built OCR pipelines using DB, CRNN, and SLANet for automated parsing of documents and designed scalable ML pipelines with API-based training and real-time processing. Developed real-time face recognition using MTCNN and ArcFace, optimizing latency from 1.2s to 0.09ms.

Devised pipelines to collect, preprocess, and label sewing-line camera data for garment worker action recognition. Implemented object detection (RetinaNet, Mask R-CNN, YOLO) and tracking (Deep SORT) for real-time analysis.

Researched model conversion to ONNX/NCNN and pruning for faster inference without losing accuracy. Successfully deployed face recognition systems on servers and Raspberry Pi, ensuring scalability.

Grade: 2.97

Completed a Bachelor of Science in Electronics and Communication Engineering. The thesis focused on non-invasive biomedical image classification and segmentation using deep learning methods.