Kafka Stream Processing Pipeline (Data Engineering)

Engineered a real-time IoT data pipeline using Apache Kafka, Python, Docker, and PostgreSQL to support reliable sensor data ingestion and processing. Built producer-consumer workflows for fault-tolerant stream processing, strengthening the data foundation for downstream analytics and ML use cases.

Refund Item Classification System (Deep Learning & MLOps)

Built an end-to-end image classification system for refund items using Python, PyTorch, FastAPI, Docker, and SQLite. Integrated API serving and automated batch processing with cron jobs to support scalable deployment and operational ML workflows. Achieved 92.35% test accuracy on the refund item classification model.

Age, Gender & Expression Recognition Mobile Application (Edge AI)

Built an Android application for real-time facial analysis using a multi-head ResNet-18 model deployed with TorchScript. Achieved 68.5% age accuracy, 90.9% gender accuracy, and 67.4% emotion accuracy through quantization and edge deployment optimization.

Vehicle Detection & Speed Estimation (Computer Vision)

Developed a computer vision system for vehicle detection and speed estimation from video using YOLO and Faster R-CNN. Evaluated model selection and real-time performance trade-offs to improve suitability for applied traffic monitoring scenarios.