Intent Detection :• Built a multilabel classifier automating intent detection in customer care call transcripts, reducing escalations by 35%, resulting in increased operational efficiency and cost savings.• Used Bert Topic clustering technique to group calls with similar intents and tagged labels for calls.• Streamlined an ensemble model in AWS Sagemaker that integrates Bert and Setfit. * Fintuned LLM model like Bloom, Flan T5 for summarizing the calls.

Automated Power Pole Inspection platform:• Built an computer vision platform for automated Power Pole inspections reducing inspection time by 25%. Used models such as Unet, mrcnn for image segmentation.• Designed and lead the data pipeline setup and model development activity for Instance and SemanticSegmentation , used PPM for precise dimension measurements.• Accelerated the model throughput by reducing the latency by 30% without affecting the accuracy by applying post-training quantization. • Used TensorRT for inference optimization and Flask along with Gunicorn to serve themodel across users in a web application.