Sujoy Datta Choudhury

Sujoy Datta Choudhury is a systems architect who has spent 25 years turning raw computation into operational intelligence—building platforms that don’t merely move data, but translate it into decisions that stand up to scale, regulation, and real-world failure modes. His career traces a consistent arc: foundational engineering in caching and web protocol standards, followed by enterprise platform leadership in finance and compliance, and now, AI-driven orchestration systems built for global throughput. Across those chapters, a single design signature repeats—contract-first interfaces, event-driven workflows, explicit failure domains, and auditability engineered into the core rather than patched in later.

At Amazon, Choudhury operates where intelligence meets high-frequency production systems. He architected the Trading Goal Orchestrator, an event-driven platform that converts publisher spend commitments into executable trading signals across 34 countries via the Commitment Hub. Built on AWS primitives (including message queues, serverless compute, container runtime, and persistent storage), the architecture is defined by idempotent processing, clear retry semantics, and durable audit trails. The system fans out across multiple processors—persisting deliverables for low-latency query and accountability, emitting pacing signals with controlled backoff, and generating bidding instructions as reusable building blocks for commitment-backed products.

In parallel, Choudhury co-architected an AI-powered keyword generation pipeline that replaced coarse page-level targeting with page- and product-specific precision matching. Using Amazon Bedrock, the system ingests URL and product signals, orchestrates prompt construction, invokes compact LLMs using token-aware batching, applies quality scoring, and persists outputs for downstream ad systems. The technical story is matched by measurable outcomes: sustained high event throughput, large-scale keyword pair generation, cost control through selective regeneration and caching, and a validated improvement in fill rate through A/B testing. In another mission-critical domain, he owns Krypton, a real-time OpenRTB enrichment service for Native Commerce Ads and ASR-C, where performance is constrained by strict bidder latency SLAs. His reliability work introduced distributed caching, tuned timeouts, graceful degradation controls, and observability patterns—decorators for targeting and creative rule enforcement, plus cardinality-aware metrics and tracing that reduced mean time to diagnosis and improved partner stability under load.

Before Amazon, Choudhury spent years at BlackRock solving a different class of scaling problem: how to operationalize regulatory complexity at enterprise throughput without sacrificing governance. As Lead Engineer and VP, he architected AFiRM, a policy-driven orchestration platform that replaced manual filing workflows with deadline-aware scheduling and transaction-level state machines. The design enabled end-to-end lineage—from source validation through filing and regulator acknowledgement—while allowing Compliance teams to evolve reporting rules through audited configuration with approvals, versioning, and validation. The operational results were decisive: high on-time filing performance, elimination of daily manual processing for a dedicated analyst team, near-elimination of false duplicates, and complete lineage coverage for filed transactions.

He also led ARGOS, a surveillance platform that replaced ad hoc SQL investigations with configurable detectors executing complex predicates over the data lake. When anomalies were detected, the platform automatically opened cases, attached evidence, and routed them through controlled workflows with four-eyes review and immutable history—introducing consistency and auditability to the operational reality of market surveillance. In IBMS, he contributed to enterprise controls designed to prevent insider trading by codifying material non-public information governance into the system itself—integrating with trading platforms and enforcing recertification, controlled cleansing, and barrier-group restrictions at scale.

Choudhury’s work bridges practice and research. His published academic work examined predictive modeling for nonprofit organizational health, comparing classical ML approaches and validating results statistically—an extension of his professional habit: evidence-driven design, measurable outcomes, and controlled experimentation. Across roles, he has mentored engineers on distributed systems, state-machine architectures, and production LLM patterns—amplifying his impact through the capabilities of the teams he builds.

The throughline is clear: security, privacy, and auditability are not optional attributes in his systems—they are architectural constraints. Whether enabling global ad trading signals or building regulatory filing platforms for one of the world’s largest asset managers, Choudhury engineers for correctness under pressure, transparency under scrutiny, and resilience under failure.