Roshan Rajan

Roshan Rajan

Software engineer with two years building large-scale recommendation and content management systems at Amazon, and leading the pilot for agentic infrastructure across the Books organization. Looking for health tech work where I own a product end to end, including getting it in front of customers.

Experience

Software Development Engineer · Amazon

July 2024 - Present

Seattle, WA

BRIX Layer: page framework for the Amazon Books retail site

  • Replaced a legacy editorial pipeline with a vendor CMS ahead of a forced ownership transfer, owning requirements for a non-technical editorial team and translating them into the criteria that decided between 3 platforms; collapsed four-team incident response to one owner and lifted traffic 35% with a redesign that federated content across the retail site.
  • Built an event-driven Slack connector that ingested team channels into an internal LLM knowledge platform, with multi-tenant auth and signed requests; the pattern was adopted by 50%+ of Books teams for ad-hoc knowledge pipelines.
  • Authored the team's AI knowledge-infrastructure strategy and led it into two workflows: an on-call agent built from its SOPs and skills, sharpened by a companion skill mining developer chat sessions, that cut on-call load 60%; and an AI page-authoring path client teams and product managers used to ship pages with no hand-written code, cutting onboarding from 2 weeks to about a day.
  • Sole owner of 2 customer-facing APIs and 15+ services: ran the on-call rotation, drove security patching and tech-debt remediation, onboarded 20+ partner teams, and cut tickets from 45 to 6 per year.

Unified Library Index: recommendations and search platform

  • Designed a zero-downtime index rebuild (parallel cluster, live cutover, audit backfill, data-quality gates), ran it in the NA and FE regions to lift data quality 30% and unblock 10+ client launches, then automated the 6-10 engineer-day process in Step Functions, reclaiming 32+ engineer-days/year across 4 regions.
  • Expanded a catalog live-update listener to 10 EU marketplaces, moving series recommendations from day-old data to real-time by fanning out events the upstream emitted only in the US; extending item-level updates across marketplaces drove a step change in write load, absorbed by a Batch Update API landing them across marketplace OpenSearch clusters at 2K requests/sec with graceful partial-failure handling.
  • Built the Recommendations Playground, letting clients create, test, and productionize strategies with no code changes, cutting implementation from weeks to minutes.

Trusted Voices: books detail-page customer experience

  • Designed the end-to-end system marketing teams used to attach and display blurbs, badges, and short-form content with products, serving 50K items to 1M+ users daily.
  • Shipped a validation and preview service for the editorial pipeline (validation API plus a UI verifying cover art and author pre-upload), preventing bad-data uploads from reaching production.
  • Automated a monthly content sync across catalog and content datastores, reclaiming 2 weeks/year and lifting stored data quality 30%.

Goodreads Auth: identity, sign-in, and account-verification systems

  • Re-architected sign-in to move 25M+ users off a third-party OAuth provider onto first-party credentials, resolving the authentication paths a customer could arrive on where failure meant permanent lockout.

Education

University of Texas at Austin

Aug 2020 - May 2024

B.S. in Electrical and Computer Engineering

GPA 3.87 · Austin, TX

Skills

LanguagesJava · Python · TypeScript/JavaScript · SQL
FrameworksSpring · React · Next.js · GraphQL · REST · Dagger
AILLM agents · RAG · MCP · Bedrock Knowledge Bases · PyTorch
DataOpenSearch · DynamoDB · PostgreSQL · Redis · Snowflake · Databricks
InfrastructureSpark · dbt · Fivetran · Airflow · Lambda · Step Functions · AWS CDK · CI/CD

Press ⌘P to save this page as a PDF.