Amit Kumar

About Amit Kumar

I\'m passionate about making AI work for good. Imagine AI that not only solves problems but does so ethically and responsibly. That\'s what I build. I dive deep into the world of Generative AI and Natural Language Processing, crafting solutions that are both innovative and trustworthy.I love to dissect complex challenges, find the root causes, and build smart, data-driven systems that make a real difference. Think of me as a problem-solver who speaks the language of AI. I\'m skilled in everything from machine learning and deep learning to the nitty-gritty of statistical analysis and data science. I use tools like Python, TensorFlow, and PyTorch to bring ideas to life.My goal is to create AI that\'s not just powerful, but also beneficial for everyone.Philosophy -> Strategic problem-solving begins with identifying the fundamental issue. I believe true progress comes from understanding the \'why\' behind the \'what. and aiming for \'how\'.Skills:Generative AI | Natural Language Processing | Machine Learning | Deep Learning | Data Analysis | Hypothesis testing | Probability distributions | Feature Engineering | Feature Selection | Python | Tensorflow | PyTorch | Keras | Flask| R | C#| SQL | Tableau | Minitab | Six Sigma | Hadoop | Agile development | Azure | AWS

Experience

1. Senior Software Engineer

   Google

   Dec 2025 — Present

   Core Impact Areas**• Performance Engineering: Optimizing LLM throughput and latency using MaxText, JetStream, and vLLM.• Accelerator Hardware: Deep expertise in TPU generations (training/inference) and TPU kernel optimization.• Distributed Systems: Architecting scalable workloads using Ray, Pathways, and JAX for multi-node efficiency.• Benchmarking & Analysis: Establishing rigorous performance baselines for LLMs to identify bottlenecks in the compute stack.**Technical Stack**• Frameworks: JAX, Flax (NNX), PyTorch (Torchview, Torch-XLA), TensorFlow.• Inference Engines: JetStream, vLLM, MaxText, TensorRT-LLM.• Infrastructure & Scaling: Ray, Pathways, Kubernetes, Orbax (checkpointing).• Optimization: Custom Kernel Development, Quantization, FlashAttention, PagedAttention, XLA Compiler.

Education

• Indian Institute of Technology, Delhi

  Master of Technology (M.Tech.), Production Engineering

  2014 — 2016

• Deenbandhu Chhotu Ram University of Science & Technology, Murthal(Sonepat)

  Bachelor of Technology (B.Tech.), Mechaincal Engineering