VASU RAJJAIN

Senior Software Engineer

JAVAPYTHONTYPESCRIPTSQLPOSTGRESQLDYNAMODBREDISELASTICSEARCHAWSBEDROCKLAMBDASTEP FUNCTIONSS3SQSDOCKERCI/CDLINUXDISTRIBUTED SYSTEMSOPENRTBRAGMCPAGENTIC AI

ABOUTCAREER & LIFE

Lead Software Development Engineer at Amazon architecting ad-serving infrastructure generating billions in annual revenue across Prime Video, live sports and DOOH. Forbes Technology Council member and IAB Tech Lab working group contributor.

EDUCATION
2017 — 2019

Master's degree, Computer Science

University of Florida
  • Grade: 3.6/4.00
  • Built CPR.VR, a healthcare VR training application using computer vision.
2016 — 2017

Senior Certificate, Computer Science

University of Florida
  • Grade: 4.00/4.00
JUN 2014 — JUN 2015

Coordinator

Sports Club
    WORK
    JAN 2025 — PRESENT

    Senior Software Development Engineer

    Amazon · Seattle, WA
    • Leading architecture and development of large-scale ad serving infrastructure generating billions in annual revenue across Prime Video, live sports and DOOH platforms.
    • Lead 30 engineers directly and influence 150+ across the organization.
    • Designed hybrid multi-tenant infrastructure handling millions of requests per second.
    • Developed agentic AI solutions using MCP servers and RAG principles.
    • Achieved 300% improvement in release management through automation.
    • Designed a BDD testing framework reducing test creation from days to under one hour.
    • Shaping industry standards through IAB Tech Lab working groups.
    FEB 2022 — OCT 2025

    Software Development Engineer II

    Amazon · Seattle, WA
    • Designed end-to-end programmatic deals infrastructure (Preferred Deals, Programmatic Guaranteed) with sub-2ms matching at 700K TPS across all Amazon publisher properties worldwide.
    • Invented Unified Language Targeting: four-signal language detection enabling multi-language ad serving across 7 streaming properties; monetized 43.6M Spanish and 95.1M French streaming hours.
    • Sole architect of a CI/CD transformation redesigning 5 pipelines — 300% deployment frequency increase, 94% fewer manual interventions, $800K/year savings, adopted by 30+ teams.
    • Principal architect of a multi-tenant platform (hybrid two-tier architecture) reducing onboarding from 52 SDE-days to 2 days, enabling billions in revenue across Prime Video, Twitch, Fire TV, IMDb.
    • Patented (pending) Reusable Ads for live events: pre-decisioning architecture serving 18M concurrent viewers at 99.945% response rate.
    • Optimized ad server infrastructure contributing to $15M+ in annual cost savings.
    DEC 2019 — SEP 2021

    Software Development Engineer

    Amazon Prime Video · Seattle, WA
    • Architected an authoritative training data store platform serving as the canonical feature repository for ML models across personalization and title classification at petabyte scale.
    • Designed and deployed a real-time copyright detection system (CAPS) using video fingerprinting and KNN similarity search across Netflix, Hulu, Disney+ and HBO content.
    • Built a distributed message deduplication service ensuring exactly-once semantics for training data propagation.
    • Developed end-to-end ML inference infrastructure for automated maturity rating generation.
    • Built scalable content classification model serving using computer vision across distributed GPU inference pipelines.
    • Designed QuickEval, an ML evaluation platform for 10 production models and 80+ model versions.
    JAN 2019 — DEC 2019

    Software Engineer

    Barclays Investment Bank · New York, NY
    • Engineered pricing computation nodes in a distributed real-time price engine generating sub-millisecond bond valuations for Fixed Income instruments.
    • Architected a high-throughput middleware messaging layer between the price engine graph and electronic trading portal.
    • Designed an automated graph cycle detection algorithm eliminating runtime circular dependency failures.
    MAY 2018 — AUG 2018

    Software Engineering Intern

    Barclays Investment Bank · New York, NY
    • Built a multi-session low-latency trading program using TCP/IP sockets and multithreading.
    • Reduced file parsing time from 2000ms to 100ms with client-side JavaScript parsers.
    • Designed a simple electronic trading system that loads different messages and log files.
    JAN 2016 — DEC 2017

    Founder

    Streetingo · Delhi, India
    • Founded Streetingo, a hyper-local hygienic street food discovery app connecting users with nearby vendors and their hygiene and taste ratings.
    • Selected among the top 200 startups for incubation by iCreate.
    • Featured in Hindi and national newspapers.
    DEC 2015 — DEC 2016

    Android Developer

    Appin Technologies
    • Worked on advanced Android development and completed a project under mentor guidance.
    • Performed testing on desktop applications, removing bugs and runtime errors from API calls.
    • Implemented revised XML layouts and API calls with 50% more efficiency.
    MAY 2015 — DEC 2015

    Junior Software Developer

    Imagine World Pvt Ltd · Bihar, India
    • Built the backend for an e-commerce website using PHP and MySQL.
    • Designed and implemented the database.
    • Worked with geolocation to render markers on maps from database data.

    APPERANCES

    AUG 2026

    The Agentic Quality Podcast

    PODCAST GUEST · CONTEXTQA

    Walked through how a team tests AI systems where the same input can produce a different output every time. Instead of validating fixed outputs, the approach shifted to testing behavior — whether the right services were called, the right rules were followed, and the system behaved correctly even when the final result changed. Deployment velocity grew from roughly 4 per month to 4 per week, eventually reaching 1,000+ deployments in a year with no manual intervention. The episode also covers agents in production, debugging with agents, human-in-the-loop decisions, trust boundaries and observability.

    2026
    MAY 2026

    PAAIS 2026 Speaker Spotlight

    PRESS MENTION · PAN AFRICAN AI SUMMIT

    Featured in the summit's official speaker spotlight covering the PAAIS 2026 Hack-AI-Thon jury. The writeup highlights deep technical mastery of distributed systems and global AI infrastructure, and evaluating AI prototypes and scalable system architectures for architectural resilience, real-world scalability, algorithmic efficiency and commercial viability.

    2026
    APR 2026

    Tenant-Aware Infrastructure for AI Systems

    WEBINAR CO-HOST · PACKT

    Co-hosted a free live session on designing AI/ML systems that serve multiple customers safely. Covered the tenant-aware blueprint, the scaling pitfalls that architectural mistakes introduce before anyone notices them, and a practical production roadmap teams can take back on day one. April 29, 2026.

    2026

    RESEARCH & BLOGS

    01
    01
    ENGINEERING · 2026

    Hybrid Multi-Tenant Architecture for Stateful Services on AWS

    Published a deep dive on rearchitecting ad-serving infrastructure away from per-tenant AWS accounts toward a hybrid multi-tenant model. Tenant onboarding dropped from 52 days to 7, infrastructure setup steps fell 80%, feature release time went from 3 days to 1, and the design scales to 100 tenants per account while preserving cluster-level isolation.

    READ POST ↗
    02
    02
    AI · 2026

    The Hidden Risks of AI-Accelerated Product Development

    Contributed perspective on what compressed timelines quietly break: pressure-testing assumptions, documenting decisions thoroughly, and keeping oversight pace with execution. Published alongside senior technology and engineering leaders from Citi, Apple, Target, Walmart, Siemens and others.

    READ POST ↗
    03
    03
    RESEARCH · 2026

    The Universe of Universes

    Co-authored with Danielle Franklin at Humanity + AI, Inc. — the organization's first published paper. The Universe of Universes (UoU) framework examines how multiple large language models perform together as an ensemble, and when simply adding more of them begins to hurt. Introduces the Benefit Yield Function (BYF) for measuring the marginal performance gained from another model, the Implosion Threshold (θ*) where ensemble quality starts to degrade, Epistemic Hereditary Drift (EHD) modeling how hallucinations and biases propagate across interconnected model lineages, and Manufacturing Velocity as a factor accelerating diminishing returns.

    READ POST ↗
    04
    04
    RESEARCH · 2026

    AGI & ASI Demystified 2026

    Co-authored 20-page position paper from SAFE AI Foundation USA that decouples intelligence from safety — a first in the field. Synthesizes 6 competing AGI definitions into a unified framework and introduces a 4-Domain Benchmark Taxonomy: Human Tasks (T), Knowledge (K), Strategy (S), and Supporting Abilities (A). Formalizes AGI as T ∧ K ∧ S, and SAFE AGI as AGI ∧ A (controllability, authorized instruction-following, safety compliance). Introduces the AGI Capability Function (ACF), AGI Criterion (AC), and AGI Capability Index (ACI) as quantitative measures. Evaluates frontier models (Claude Opus, GPT-5.5, Gemini 2.5 Pro, Grok 4) against the taxonomy. Advocates for a "Safe AI Operating Point" to prevent catastrophic risk while preserving economic benefit.

    READ POST ↗