2025-11-09 –, Room 121
AI/ML workloads depend heavily on complex software stacks, including numerical computing libraries (SciPy, NumPy), deep learning frameworks (PyTorch, TensorFlow), and specialized toolchains (CUDA, cuDNN). However, integrating these dependencies into Bazel-based workflows remains challenging due to compatibility issues, dependency resolution, and performance optimization. This session explores the process of creating and maintaining Bazel packages for key AI/ML libraries, ensuring reproducibility, performance, and ease of use for researchers and engineers.
Introduction to Bazel for AI/ML (20 min): The importance of Bazel for AI/ML workloads and why it’s useful for dependency management.
Challenges in AI/ML Bazel Packaging (30 min): Challenges in building Bazel packages for AI/ML libraries, including transitive dependencies, build system differences, and GPU acceleration.
Strategies for Packaging (30 min): Strategies for packaging libraries like SciPy, PyTorch, TensorFlow, and other dependencies while ensuring compatibility and performance.
Best Practices for Distribution and Maintenance (20 min): Best practices for maintaining and distributing Bazel packages for AI/ML.
Hands-on Demo (144 min): A demo of building and using AI/ML libraries with Bazel.
Q&A and Open Discussion (36 min): Open discussion and questions.
Ramesh Oswal is a Senior Motion Planning Engineer at Aurora, with experience from Luminar and Noble.AI. He has expertise in AI/ML for Autonomous Systems and Education. He has also served as a review committee member for NeurIPS 2024, CNCF 2024, and CNCF 2023.
Jiten Oswal is an engineering leader and AI systems architect based in San Francisco with over 14 years of experience building large-scale data and AI infrastructure. He spent several years at Salesforce, earning 6 U.S. patents in cloud data and AI systems. Later, he founded and led an intelligent automation startup as CTO. Currently, he’s the founding AI Lead Engineer at a AI Startup developing an open-source AI Agent Platform for Enterprises.