Akash Gupta

Developed and optimized deep learning models for segmentation, classification, and anomaly detection on medical imaging data, including large-scale Whole Slide Images (WSIs). Designed and deployed Vision-Language Models (VLMs) for multimodal question answering on medical images. Built multimodal pipelines integrating image encoders and LLMs to process visual and textual inputs.

Conducting research on optimization of LLM inference pipelines, targeting reduced latency and improved throughput. Surveying methods related to model optimization (quantization, pruning) and KV cache management strategies.

Developed and optimized ML models for precise classification of giga-pixel images. Achieved superior accuracy and computational efficiency in giga-pixel image classification using Patch Gradient Descent.

Developed innovative training pipelines for CNNs and Vision Transformers. Implemented and benchmarked quantization methods in TrAILMet, an open-source PyTorch library for model compression. Conducted large-scale experiments on model compression and optimization techniques.