Jubin Thykattil

About Jubin Thykattil

Software Development Engineer at Amazon with nearly 6 years of experience leading the design and delivery of distributed systems in e-commerce and B2B advertising, with a growing focus on Gen-AI and AI agent architectures.I lead complex, cross-functional initiatives from ambiguous problem spaces to production — designing systems that scale globally, making deliberate technology tradeoffs, and enabling teams to move independently. Most recently, I\'ve been leveraging Gen-AI to build production systems that automate content creation for marketers — designing multi-agent architectures where agents are scoped to specific use-cases, orchestrated to handle end-to-end workflows, and integrated into existing platforms to dramatically reduce the time marketers spend on manual content production. My broader work spans real-time caching architectures with global replication, data pipeline modernization using Airflow and AWS Batch, and high-throughput event-driven services built on SNS FIFO and Fargate.I\'m equally invested in the people and processes around the code. I\'ve redesigned org-wide operational review processes for 15+ teams, mentored engineers across levels from L4 to L6, run a cross-org book club on software design, and serve as a Privacy Bar Raiser. I\'ve conducted 85+ technical interviews and led tech survey analysis across 134 respondents to turn employee feedback into actionable engineering improvements.I\'m most energized by the convergence of Gen-AI, agent design, and large-scale distributed systems — where the challenge isn\'t just making AI work, but making it work reliably, cost-effectively, and at scale within production infrastructure.

Experience

1. Software Development Engineer - Amazon Business Marketing Tech

   Amazon

   May 2024 — Present

   Led and launched AI marketer for Amazon Business • Led a cross-functional team of four engineers and designed an AI-powered agentic platform that automates onsite marketing campaign creation across 10+ global marketplaces, replacing a multi-tool manual process with 3-day to 4-week SLAs. • Built the core Content Agent on Amazon Bedrock Agents with a hybrid prompting strategy combining chain-of-thought reasoning and dynamic one-shot prompting. • Architected the knowledge base on Amazon Kendra with metadata-filtered semantic search to reduce hallucination. • Integrated Bedrock\'s LLM-as-a-judge for iterative content quality evaluation. • Designed persistence on DynamoDB with an OLAP pipeline for reinforcement learning. Led and launched Paid Search for Google Text ads for Amazon Business • Led a team of five SDEs and collaborating with data engineers and applied scientists to improve Amazon business text ad relevance on Google, targeting $14M-$20M in ad spend savings, and $1M in Adobe ad tech fees.• Successfully deployed offline machine learning models on AWS, leveraging Docker containers, Amazon ECR for secure model storage, and AWS Batch for scalable execution, with automated CI/CD pipelines.• Orchestrated big data extraction workflows using AWS Step Functions, Lambda, and S3 triggers. Automated the process of gathering customer ad click data from multiple data lakes across Amazon.• Utilized AWS CloudWatch for real-time monitoring of application performance and health. Established SOPs and best practices with Science and Data teams, introducing observability and CI/CD integration, resulting in more reliable and maintainable ML pipelines.• Supported in hiring, mentoring and transitioning this charter to a new team

Education

• Dwarkadas J. Sanghvi College of Engineering

  Bachelor’s Degree, Electrical, Electronics and Communications Engineering

  2013 — 2017

• North Carolina State University

  Master's degree, Computer Systems Networking and Telecommunications

  2017 — 2019