At Genesys International Ltd., I fine-tuned point-cloud semantic segmentation models for utility-pole and conductor classification from aerial LiDAR, reaching 0.87 mIoU overall and 0.93 recall on the pole class.
I also built a Python and Flask data pipeline that converts 3D measurements into validated training data, cutting annotation effort by ~70%. The work connected labelling and retraining in a production ML workflow.
For my enterprise RAG project, I built a FAISS retrieval pipeline with Google Gemini 2.5 Flash so stakeholders could query 50K+ structured fintech records in natural language instead of writing SQL. The system reduced insight discovery time by ~60%.
My projects also include YOLOv11 melanoma detection and federated learning. I work across applied machine learning, computer vision, and LLM systems, from data preparation and evaluation through application integration and deployment.
