Amazon Senior Software Development Engineer
2021 — PRESENT
Java · TypeScript · Python · GraphQL · React · CDK · AWS · Bedrock · Agentic Systems · RAG
L4 → L5 · OCT 2023 / L5 → L6 · APR 2026
Platform Architecture & Distributed Systems
- Led the technical vision, cross-org alignment and delivery of the Case Management platform for Amazon's entire Customer Service organization, unifying all 49 touchpoints handling 1B+ annual interactions. Secured Principal-Engineer alignment across 5+ orgs — eliminating 991K repeat contacts/year ($11.15M) and cutting worldwide Average Handle Time by >10s ($14M annualized).
- Architected a large-scale, event-driven workflow orchestration platform at 99.99% reliability, establishing a reusable integration model adopted by 15+ teams and cutting onboarding time from 3 weeks to <1 week.
Agentic & Generative AI
- Architected a net-new AI commitment lifecycle that extracts, verifies and policy-validates promises from customer conversations, then tracks them through SLA-based escalation and backend-verified closure — targeting a gap of ~38M unverified promises and 10M+ customer re-contacts annually, representing ~$3.4B in estimated customer-spend exposure.
- Conceived and built an AI-native engineering execution system that turns project-level intent into persistent dependency graphs; designed the orchestration model, agent execution lifecycle, dynamic replanning and human safety boundaries enabling parallel agents to carry coding, research, review and downstream work across 10+ real engineering initiatives and dozens of task / CR lifecycles.
- Built and operated an always-on agentic on-call system for a live Amazon production queue; designed the architecture for autonomous triage and investigation across logs, metrics, code, deployments and internal knowledge, with durable incident state and deterministic safety gates across 10 weekly rotations and 80+ real tickets.
- Shipped a natural-language analytics product pairing multi-stage LLM reasoning with semantic retrieval over 440GB+ of operational data, cutting time-to-insight from 15 minutes to 30s and removing a standing engineer dependency for reporting.
- Automated a recurring engineering operations workflow with LLM agents + human approval gates, cutting average resolution time by 80.9% (137h → 26h) with 0% workflow errors; presented org-wide as a reusable HITL pattern.
- Scaled a Bedrock context-summarization pipeline to multi-millions of summaries/month at sub-second latency, reducing associate comprehension time 35%; complementary visibility tools served 200+ users and cut data lookup time 30×.