Zoox

Uber does have dozens of partnerships with all of these other autonomous vehicle firms, and we haven't really seen them because a lot of them haven't come into fruition yet.

Disaggregated Inference, Part 3: Why Your Networking Stack May Not Be Ready ( https://lnkd.in/dCpfsFTk ) Last week I posted about how the KV cache actually moves between prefill and decode. Once you've decided to send it, the next problem shows up fast: what network stack can move gigabytes between GPUs in milliseconds? A few moves that matter: ✳️ NCCL burns GPU streaming multiprocessors moving bytes. On the 256KB–1MB messages typical of KV transfers, that costs 30–50% throughput vs purpose-built P2P engines. ✳️ Three purpose-built transports have emerged, each solving a different problem. NIXL abstracts heterogeneous memory tiers; UCCL does P2P without SM overhead and runs across NVIDIA and AMD; Mooncake's Transfer Engine aggregates across NICs to hit 190 GB/s on 8×400 Gbps. ✳️ AWS and Cerebras announced a collaboration to ship a solution on AWS Bedrock that will run prefill on AWS Trainium and decode on the Cerebras CS-3, stitched together over Elastic Fabric Adapter. Cross-vendor disaggregation isn't research anymore. The punchline: when the chips on either side of the wire are built by different companies, the data plane stops being plumbing and becomes the product. That wraps the 3-part series. Earlier posts: Part 1: When prefill/decode disaggregation is worth it: https://lnkd.in/dwiRQKUz Part 2: Moving the KV cache without stalling decode: https://lnkd.in/dpjp6b_F #LLMInference #AIInfrastructure #KVCache

One is enterprise owned fleet. The other one is event-based campaign.

They got approval from the NHTSA to run their like really, really innovative and cool robot vehicles as well.

I’ve joined the Strategic Partnerships team at Waymo . I'm really excited to work with Shana Downs , Madeline Csere , and a few other familiar faces 🙂 I've long believed that AVs will meaningfully increase access to safe, accessible, and equitable transportation, while decreasing carbon emissions - thrilled to play a small part in bringing this to more communities. Photo is Jay Kornder II and me at the 2018 New York Auto Show, where the Waymo Jaguar I-pace was unveiled!

I designed this drivetrain/differential concept in Shapr3D using VR to explore component interaction, packaging, and immersive engineering visualization beyond traditional CAD environments. It’s exciting to see how VR tools could transform future engineering workflows, product development, and design reviews. #CAD #VR #Shapr3D #MechanicalEngineering #AutomotiveEngineering

I'll be speaking at ClickHouse Open House in San Francisco (May 26–28) about how we use ClickHouse / ClickStack for Observability at Zoox , and what we learned while adopting it at scale. If you're using ClickHouse for observability, I'd love to connect! https://lnkd.in/gnMx7FEn

I'm presenting a poster at PyCon 2026 🐍 "How Python Powers Modern AI: Comparing PyTorch and JAX Execution Stacks" One question drove the whole thing: when you call torch.matmul (a, b), does it hit cuBLAS directly — or does TorchInductor compile a fused Triton kernel? Spoiler: it depends on whether torch.compile is in the picture. Come chat at the poster session if you're around. Always happy to dig into inference optimization. #PyConUS #PyCon #Python #PyTorch #JAX

Pair your phone once, auto-connect for every ride — the best music experience in the whole wide world of mobility. Super proud to have contributed to the content design & product marketing for this. Congrats Josh Redel Maria Jose Perez Calvo Joshua Yao Thibault de Waziers Alex Wolowiecki Loarna Hassen May Wong Monique M. and so many others!