Hi, I'm
Babu Sovereign
Perumalla
Technical Lead · San Francisco Bay Area
I build systems that move money — and AI agents careful enough to be trusted near it.
I’ve spent my career building the systems big retailers run on: the pipeline that tells a grocer what two thousand stores are doing right now, the gateway that makes sure nobody gets charged twice, the services behind the catalog you browsed last night.
Payments taught me a particular kind of carefulness — when a bug is somebody's money, "mostly correct" isn't a thing. These days I bring that same carefulness to AI agents: real tools, real guardrails, and real evals before I'd trust one with real work.
What I do
Four threads, woven together.
Four threads of one career, woven like the pattern on a banknote — engravers use it because it's hard to fake. Click to stir the loom.
Distributed backends
Home base — services that stay up through holiday peak, and APIs other teams actually enjoy consuming. Mostly Java, Spring Boot, Kafka and GraphQL.
Streaming data
Raw store activity turned into numbers a business can act on in minutes instead of overnight. Databricks, Scala, PySpark, Python, and a lot of SQL on GCP.
Payment Systems
My favorite kind of hard: a duplicate request here is a double charge. Tokenization, refunds, idempotency, audit trails — correctness first, always.
AI agents
The newest thread, growing fastest. They’re my daily backbone for backend work — building APIs, testing, reviews and documentation — cutting the manual effort and shipping features with real context. Claude API, MCP, Claude Code, Python.
The ledger
Seasoned in retail, called on for the hard problems.
All of it as a technical lead at Nisum Technologies, embedded with retailers you'd recognize. It's client work, so I'll spare the names here — happy to go deeper over coffee.
Real-time store metrics for a national grocer
Kafka into Databricks, served over GraphQL — live operational numbers for 2,000+ stores, rolled up from store to district to division.
Card tokenization in an online payment gateway
Tokenized the whole purchase path so no service held a real card number.
Point-of-sale and payment operations
Store POS transactions — purchases, refunds, receipts — and the tools agents use to issue card refunds and account credit. I handled payment operations and receipt processing, end to end.
Three years inside a national department store
Holiday-peak stability work — including a datastore migration under live traffic — then pricing, promotions, registry and personalization.
Fulfillment migration for an outdoor retailer
Legacy warehouse and order management, re-platformed onto GCP microservices with MongoDB.
Search and product discovery for a multi-brand retailer
The services behind search, catalog feeds, inspiration boards and AI-powered related searches, across several well-known brands. Java and Spring Boot on Kubernetes, with Cassandra underneath.
The unglamorous tools that keep retail running
Photography workflow systems for product imagery at one apparel giant; demand-forecast extracts that decide store replenishment at the same one, years earlier. I built the extract architecture and the delivery pipeline around it.
Away from work
I built a fortune cookie.
Morsel Jar is a free iPhone app I designed and happily vibecoded with my AI agent — one small encouraging morsel a day, in the tone you pick, about the part of life you care about right now. It's the opposite of everything I build at work: no backend, no account, no data leaving your phone. Just SwiftUI, a jar of hand-written fortunes, Siri recall and a widget.
"The app doesn't need a brain. It needs a heart."
That was my one design rule. Everything lives on your phone — the app can't phone home because there's no home to phone. Coming soon to the App Store.
These are the engineer's edition, written for this page — the ones in the app are kinder.
Now
What I'm into at the moment.
Going deep on AI agents
Tool design, MCP, and especially evals. I already code with agents every day; now I'm building the systems side of it. I think evals are to agents what tests were to services: the thing that separates demos from systems.
Writing fortunes
Morsel Jar needs hundreds of them, hand-checked, across three tones and four focuses. Harder than it sounds; more fun than it should be.
Out and about
You'll find me at Bay Area meetups and tech events — usually in whatever conversation is about payments correctness, or about trusting agents with real work.