Aniruddh Chandratre

Distributed data platform · GPU scheduling · cluster efficiency

• Lead Tesla AI's internal data processing platform for acquisition, curation, mining, auto-labeling, training, and evaluation; it runs 10B+ jobs and 3,000+ compute-years/month across 10 GPU clusters.
• Re-architected the platform core, distributed worker coordination, and memory model, increasing throughput from 940M to 15B+ data items/month.
• Built scheduling and live defragmentation for a heterogeneous GPU fleet, increasing usable compute capacity from 24% to 88% and fleet-wide effective utilization to 98%.
• Added end-to-end lineage across artifacts, datasets, training jobs, and models for multi-stage pipeline provenance.
• Built internet-scale acquisition and curation pipelines across Tesla AI programs.

Edge-to-cloud data infrastructure for the gigafactories

• Designed and shipped an edge-to-cloud data broker processing 10B+ manufacturing data points/day across Austin, Berlin, and Fremont.
• Developed an mmap-backed durable queue sustaining 5M writes/s on NFS-backed Kubernetes volumes and a distributed cache processing 10M entries/s, cutting storage 80% across 50 deployments.
• Built and operated 20 high-availability services for factory insights.

• Researched formal verification under DARPA ARCOS with Lockheed Martin; co-authored PSY-TaLiRo and operated the Kubernetes cluster for high-volume simulation workloads.

• Built the image-ranking model at the core of AfterShoot's photo-culling product, now used by approximately 250K photographers.