Benchmarked and profiled LLM inference on single and multi-node (A100, H100) systems using vLLM, SGLang, and Tensorrt (w/ triton inference server) leveraging MLPerf Inference (Datacenter Suite) for offline, online, and interactive scenarios.

Identified and resolved performance bottlenecks via Nsight Systems, Nsight Compute, and PyTorch Profiler; optimized throughput and latency through quantization of weights, activations, and KV cache.

Contributed improvements and findings to MLPerf, vLLM, and SGLang GitHub repositories; authored a technical report guiding optimization strategies for production-scale LLM inference.

Doing research and development to predict hepatocellular carcinoma (HCC) development using advanced deep learning models on histology images, focusing on “time to event” analysis.

Worked primarily with Python, OpenCV and TensorFlow to preprocess and build deep learning models for survival analysis using large scale image datasets.

Presented a comprehensive machine learning project on a customer churn use case at the SAS Innovate Conference to 200+ attendees, demonstrating the capabilities of SAS solutions and their integration with open-source tools on the SAS cloud environment.

Built generative AI RFP applications using LlamaIndex and Streamlit with integrated human feedback loops, hosted on VMs with LLM monitoring for pricing, performance, and resource use—cutting processing time from weeks to hours.

Led end-to-end ML projects using synthetic data and ensemble methods, deploying with SAS Viya and open-source tools; supported client adoption through presentations and hands-on workshops focused on model management and decisioning.