Bio
Asma Beevi K T is a senior engineer at NVIDIA, developing the NVIDIA TensorRT Model Optimizer toolkit. Her interests span training and inference optimizations for deep learning models, particularly LLMs.
Sessions (2)
Additional Speaker Details
Using publicly available information we constructed an analysis to help you get a feel for this speaker before deciding to attend their session.
Asma Beevi (publishing as Asma Kuriparambil Thekkumpate) architects and leads NVIDIA's Model Optimizer library, shipping production quantization techniques — QAT, quantization-aware distillation, and NVFP4 — that recover accuracy for low-precision LLMs like Llama Nemotron, making her session directly relevant for engineers deploying large models at scale. Her recent work spans both the algorithms and the practical tooling (TensorRT Model Optimizer), grounded in a background as a founding engineer at OmniML, which NVIDIA acquired in early 2023.
Speaking style
GitHub
@realAsma- realAsma/BeeLoop★ 3Event-driven loop / personal-assistant system: a perpetually running gateway coordinating a hive of agents (BeeBot on the Claude CLI).
- realAsma/AgentInProduction★ 1Reference repo for giving coding agents project-specific guidelines, adapted from an NVIDIA Model Optimizer contribution.
- NVIDIA/Model-Optimizerorg★ 3,000Unified library of SOTA model-optimization techniques (quantization, distillation, pruning, speculative decoding) for TensorRT-LLM/TensorRT/vLLM deployment — the library Asma leads at NVIDIA.
Recent writing (5)
- Fundamental Pieces for Building Any Claw or Loop · blog · Jun 2026
- BeeLoop: My Loop That Engineers My Other Loops · blog · Jun 2026
- Quantization-Aware Distillation for NVFP4 Inference Accuracy Recovery · paper · Jan 2026
- How Quantization Aware Training Enables Low-Precision Accuracy Recovery · blog · Sep 2025
- Fine-Tuning gpt-oss for Accuracy and Performance with Quantization Aware Training · blog · Aug 2025