NeMo Framework:

• Leading the NeMo Framework's inference operations.

• Designed and implemented the export module to run fast LLM inference for community models such as Llama2, GPT, Falcon, Starcoder, and NeMotron using the TensorRT-LLM.

• Designed and implemented the deploy module to deploy any LLM model to Triton Inference Server.

GNN Tool:

• Lead the API design and the development of the GNN Tool project.

• Built GNN based models for fraud detection, recommender systems, and NLP.

• Worked with the product managers to get the requirements and built the roadmap.

• Created the design documents and wrote the initial code for the components.

• Managed the weekly team meetings, and created and assigned tasks.

• Helped engineers to complete tasks when needed and performed code reviews.

• Managed the inter-team communications.

• Presented our work to customers and helped with the technical discussion.

GPU Accelerated Recommender Systems, Merlin:

• Working on GPU accelerated feature engineering and preprocessing for recommender systems.

• Focusing on GPU accelerated Deep Learning library for recommender systems.

GPU Accelerated Machine Learning, cuML:

• Focused on traditional machine learning (ML), deep learning, and HPC problems.

• Developed GPU accelerated ML algorithms to help data scientists and researchers train ML models very fast on multi-node multi-gpu (MNMG) AI infrastructure.

• Architected and developed GPU accelerated ML primitives library for NVIDIA's open source project RAPIDS (http://rapids.ai).

• Developed single-GPU accelerated generalized linear models (OLS, Ridge, Lasso, SGD, etc), PCA and, tSVD using C++ and CUDA for RAPIDS cuML.

Achieved more than 20x speed-up compared to scikit-learn's implementation.

• Implemented MNMG linear models (OLS, Ridge, Lasso, Elastic-net, SGD, etc), tSVD, and PCA, and achieved 40x-90x speedup.

• Guiding the data scientists and researchers in building innovative solutions based on NVIDIA technology.

• Being an industry thought leader on integrating NVIDIA technology into deep learning solutions to support scientific and engineering applications.

• Giving deep learning workshops on computer vision (CV) and natural language processing (NLP) using deep neural networks (convolutional neural network, recurrent neural networks, long-short term memory networks, etc) to help developers, researchers and data scientist adopt the technology.

• Focusing on deep learning applications in financial services industry and insurance.

• Developed 3 use-cases using LSTMs, deep autoencoders, and deep reinforcement learning (policy gradients, deep Q-learning actor-critic, etc) for capital markets.

• Focusing on some of the deep learning based NLP tasks such as named entity recognition, text summarization, and topic modeling for finance use-cases.

• Developed web services/APIs using big-data technologies including Kafka, and Cassandra.

• Worked on predicting the trends of the hotel rates using simple linear regression.

• Did the proof of concept of machine learning based collaborative filtering to recommend hotels to customers.

• Focused on weighted alternating least squares (W-ALS) based recommender system.

• Taught Microprocessor Laboratory to a diverse class of 30 undergraduate students.

• Advised 100 graduate students for GPU and FPGA course projects.