I'm a software engineer at Amazon. I design and run distributed systems, and I build AI agents and the infrastructure behind them.
Selected work
- Agent tooling platform: Built and own a platform that turns recorded browser workflows into agent-native CLIs with a uniform JSON contract, self-correcting error hints, and prompt-injection defenses. It serves engineers across multiple divisions and uses substantially less model context than an equivalent MCP integration.
- Agents with production authority: Designed one of my organization's first agents allowed to act on production tickets, gating actions on confidence and reversibility, with a full-census safety evaluation that found no unsafe actions.
- Domain engineering agents: Built workflows for creating domain engineering agents for code review, code changes, on-call, and architecture questions, now used across much of the team's daily work.
- A knowledge map for agents: A knowledge map and harness that lets agents find the right service, its owners, and its context, built on the Open Knowledge Format. It automated most repeatable maintenance and ticket work.
- Distributed systems: Event-driven services that measure and enforce delivery performance for sellers using Fulfilled by Merchant, processing over a billion events a day across dozens of marketplaces. Coordinated a live cross-region migration of tens of terabytes and executed the production cutover.
- Security: Security reviewer for tier-1 applications, including threat modeling, and author of security guidance used across Amazon.
Mostly Java and Python, on AWS (DynamoDB, SQS, SNS, Lambda, CloudWatch) with Redis, OpenSearch, and Postgres. For detailed results or a resume, write to hello [at] swopnil.com.
Writing and talks
I write about what I'm building on this site. I've also spoken at internal conferences and run AI engineering workshops for my organization.
Experience
Distributed systems for seller fulfillment and performance, AI agents and developer infrastructure, and security reviews. I also mentor engineers and build developer tools used across Amazon.
Built an early conversational-AI commerce platform, from launch through its early growth.
Backend components for an NLP-based telehealth platform, in Clojure.
An email analytics tool that used social-network graphs to flag unlawful communications.
Principles
A few ideas I come back to when designing systems and working with teams:
- First principles: Start from what must be true, not from how it's usually done. Break a problem down to its fundamentals before choosing a solution.
- Second-order thinking: Ask what happens next. A change that saves time today can add cost, load, or risk somewhere else later.
- Systems thinking: Look at feedback loops and at how a change travels through connected parts. Inspired by Thinking in Systems.
- Conway's law: Systems mirror the teams that build them, so service boundaries and team boundaries should be designed together.
- Jevons paradox: When something gets cheaper, we use more of it. AI makes code cheap to write, so clear requirements, review, testing, and agreement on what a system is for matter more, not less.
Education
President of the college's ACM chapter: taught student developers and organized talks and panels.