About Me

Hi! I’m Udit Saxena.

I am a Lead Machine Learning Engineer at ASAPP, where I design large-scale AI infrastructure for Generative AI systems.

At ASAPP, I build retrieval-augmented generation (RAG) systems and multi-turn conversational agent platforms that power enterprise-scale voice and text AI products.

My work spans workflow orchestration (Airflow, Spark, Kubernetes), LLM evaluation and inference infrastructure, and distributed deep learning training. I collaborate closely with our Research, Product, Platform, and SRE teams to bridge cutting-edge LLM research with production systems.

I also spoke at Airflow Summit 2024, sharing how ASAPP orchestrates large-scale batch inference for gen-AI workloads across Kubernetes clusters (Airflow Blog).

Previously, as a Senior Machine Learning Engineer at Sumo Logic, I built fast and approximate streaming clustering algorithms for text streams and time series models, resulting in two granted patents.

My research spans natural language processing, geometric and topological deep learning, and sparse methods for model efficiency. I recently published Scalable GPU-Accelerated Euler Characteristic Curves at NeurIPS 2025 (NeurReps Workshop) — introducing optimized, differentiable topological features for PyTorch with 16–2000× GPU speedups on modern architectures.

Earlier, in A Unified Paths Perspective For Pruning at Initialization with Thomas Gebhart, we studied how neural network path structure relates to pruning using the Neural Tangent Kernel framework.

Before that, I worked with Microsoft Research, Lexalytics, UMass IESL, Google Summer of Code (MLPACK), Sprinklr, and Adobe on topics ranging from active learning to graph neural networks for text.

I enjoy the intersection of AI systems, distributed computing, and geometric ML, and I’m always interested in building tools and infrastructure that make large models more efficient, interpretable, and useful.

View my resume