Production AI, built on distributed systems thinking

Vivek's public work spans multimodal processing, computer vision, document intelligence, and foundational cloud infrastructure. The technical details below stay at a public, non-confidential level.

Amazon Bedrock Data Automation

Multimodal AI begins with a practical systems challenge: convert varied content into useful, structured information at a scale and cost enterprise workflows can support.

Bedrock Data Automation processes documents, images, audio, and video for generative AI and analytics use cases. Vivek's publicly profiled leadership spans the engineering, applied science, and product work behind this service area.

The architecture question is larger than model quality alone. Production use also depends on handling complex inputs consistently, matching infrastructure to workload, keeping latency predictable, and maintaining viable unit economics.

Explore the AWS service ↗   Read published perspective ↗

Amazon Rekognition

Computer vision platforms have to move beyond a successful model call. They need dependable pipelines for image and video analysis, sustained service availability, and regional reach close to customer workloads.

Public launch updates from Vivek describe Rekognition availability expanding into Bangkok, Kuala Lumpur, and São Paulo. Regional delivery brings the operational realities of rollout, service readiness, and customer access into the technical story.

Explore Amazon Rekognition ↗   View regional launch update ↗

Amazon Textract

Document intelligence combines recognition, layout, handwriting, forms, tables, and extraction into information that software can use. At scale, the task calls for an engineering approach that treats model output as one part of a larger workflow.

Vivek's public leadership profile places Textract alongside multimodal and vision services, connecting document processing to the shared foundations of high-throughput data extraction and reliable cloud delivery.

Explore Amazon Textract ↗

Distributed storage foundations

Earlier cloud work covered EBS Snapshots, including archive tiering and copy, and EBS Volumes. The issued patent US 11,262,918 B1 concerns reducing uneven drive wear in distributed storage.

This systems foundation is relevant to production AI because both domains must manage resource contention, uneven workload, service durability, and predictable operation across a fleet.

Engineering themes

Workload-aware infrastructureUse systems that fit actual demand, not only peak theoretical throughput.
Inference economicsBalance latency, utilization, and cost across sustained production workloads.
Reliability as a design propertyBuild recovery, verification, and operational visibility into service architecture.
Regional deliveryTreat availability, latency, and data location as part of customer experience.

These summaries describe publicly documented product areas and professional work. Proprietary architecture, customer names, and internal metrics are not included.

Follow the ideas behind the systems.

Selected media and expert commentary on production AI.

Media and commentary →