Kabir Mehta

Backend Engineer at Tallyfern Payments | APIs, payment reliability, and useful code reviews

I’m a backend engineer in Bengaluru, working on payment APIs and the systems that keep transactions moving when something inevitably times out. I like debugging messy problems, writing down what we learn, and helping junior engineers get comfortable asking questions. Most of my work is in Java, PostgreSQL, and Kafka, with a healthy suspicion of anything described as a quick retry. Outside work, I climb indoors, bake stubbornly mediocre sourdough, and disappear into the Western Ghats when the rain arrives.

Build and maintain payment orchestration and reconciliation services using Java, PostgreSQL, and Kafka. Work on idempotency, webhook delivery, and incident tooling. Mentor two junior engineers through pairing sessions and code reviews, and help maintain the team's onboarding guide.

Developed backend services for checkout, refunds, and order tracking. Improved integration tests around payment failures and helped move slow reporting queries out of customer-facing request paths.

Built REST APIs and internal support tools for inventory software. Learned production debugging, database migrations, and why a successful local test is not quite the end of the story.

Java, Spring Boot, PostgreSQL, Apache Kafka, REST APIs, Distributed Systems, Payment Integration, Observability, Technical Mentoring, Integration Testing

English, Hindi, Kannada (conversational)

A junior engineer asked why we store an idempotency key instead of just checking whether the payment already exists. We ended up drawing two simultaneous requests on the whiteboard, which did more useful work than my first ten minutes of explanation. Good reminder for me: start with the failure case, not the vocabulary. We've added that little diagram to the onboarding notes. My handwriting is now a production dependency.

Spent part of this week making a reconciliation alert less noisy. Not a new service, not a rewrite. Just separating records that genuinely needed attention from records that were still within the expected processing window. The useful bit was sitting with the operations team and watching how they investigated each alert. There were three clicks we could remove before we needed any clever engineering. Happy with this one.

Trying a small change in code reviews: if my comment needs a paragraph of explanation, I offer a ten-minute pairing session instead. We still write down the decision afterwards, but it avoids the strange situation where two people spend an afternoon politely misunderstanding each other in a thread. Not every review needs a call. Some definitely need fewer essays from me.

Updated our onboarding exercise this week. It now includes a webhook that arrives twice and one that arrives late, rather than only the very well-behaved happy path. The goal isn't to catch anyone out. It's to give new teammates a safe place to ask what happens next before they're looking at a real incident. Also removed a setup step that apparently only worked because of something installed on my laptop. Humbling.