Ashish Sardana

Master's

Artificial Intelligence and Robotics

NVIDIA

Lead Machine Learning Engineer

• Speech AI (Launchpad, Riva): o Drove execution and development of Riva (ASR & TTS) labs, from conception to launch, on Nvidia Enterprise Cloud o Conducted market research, defined user experience journey and iterated on feedback from different user personas o Grew product adoption & enterprise revenue by 20% and y% respectively by also publicizing tech marketing assets • Machine Translation (NeMo, Triton): o Explored various tokenizers, data cleaners, and transformer of sizes (100M-1B) to deliver a state-of-the-art NMT model o Increased NLP product adoption by 25% via end-to-end tutorials and advised EMEA customers with professional services • Fraud Detection (PyTorch Graph, Rapids): o Modeled tabular credit card dataset into graph and trained relational graph CNNs leading to 10% gain in classification accuracy o Demonstrated proof of concept to American Express and published a blog on optimizing fraud detection with GNN on GPUs • Semantic Search (Megatron LLM): o Designed information retrieval system architecture for Nvidia support forum to augment elastic with advanced NLP search o Implemented end-to-end neural retrievers resulting in 8% and 12% absolute gains for recall & exact match metrics respectively o Collaborated with Deepset.ai and AWS team to create use-case specific examples for joint developer outreach program • Recommendation System (GeForce Now, Merlin): o Improved game ranking accuracy by 15% using wide & deep model and reduced gamer churn by 10% for M users x K games o Produced competitive analysis of the recommender libraries & startup landscape to address slow adoption of Nvidia Merlin