Summary
Himanshu Gupta is an Applied Scientist with 18 years of experience building and deploying ML and LLM systems, currently at Amazon after recent roles accelerating production LLMs for Indic languages. He led creation of a multilingual 7B foundation model and a compact ~100k-word tokenizer, trained on nearly 2 trillion tokens and optimized for high-throughput, low-latency serving with vLLM and TensorRT. His research background from Arizona State University produced multiple ACL/EACL/NeurIPS/NAACL papers on novelty detection, numerical commonsense, instruction tuning, and long-sequence evaluation, bridging academic rigor with production impact. Prior work at American Express focused on document information extraction and graph-attention models, resulting in multiple patents. Comfortable spanning research, model engineering, and deployment, he combines deep NLP research instincts with pragmatic systems engineering in large-scale settings. Based in Bengaluru, he also experiments as a hands-on developer on GitHub, indicating a practitioner mindset that values tooling and reproducible engineering.
18 years of coding experience