I'm Director of Engineering at Omnify, where I lead a team of 12. I've spent ten years building SaaS, payments and cloud platforms. For the last two I've been building AI: LLM-based products, and agents that help my team test and document our software.
In numbers
Selected work
A documentation platform written by AI and checked by AI. It maps our apps into a code graph and builds an evidence pack for each feature. One model writes every sentence with a citation, Claude Opus checks each claim, and any page with an unsupported claim is held back. I ran a blind comparison of models to pick the writer: it beat our existing pages by 1.0–1.42 points.
An orchestrator that directs worker and reviewer agents. 14 agents documented 500+ APIs in parallel, 10–15 agents test the product in a real browser, and a nightly agent keeps the docs in step with the code.
One agent serving three products for ~10 enterprise customers: chat over live data, drill-down dashboards, licence-savings reports and what-if simulations. Each session runs in its own microVM, and each customer's data is walled off in the database itself.
I rebuilt Omnify's AWS as a multi-account setup in Terraform and cut the bill by 50%. Then customer vanity domains on CloudFront, and a multi-region email platform with per-customer sender reputation, each rolled out staging → preprod → prod.
I built the payment system that handles over $1M a month across Stripe, PayPal and Razorpay, then redesigned it on Stripe Connect with a disputes flow and a Stripe-compatible proxy for enterprise customers.
An AI product that pulls in Facebook and Instagram ads, tags every image and video with vision models, predicts which creatives will hit their targets, and explains why in plain language.
I moved a multi-tenant reporting product onto a second cloud without forking the code, then hardened it for a regulated enterprise client until it passed their security review.
Agents at Omnify
I split the documentation of our platform across 14 agents working in parallel, with reviewer agents checking their output. Scroll to see it run.
Model evaluation
To choose the model that writes Docs Hub pages, two judge models from different providers scored each candidate blind. The AI-written pages beat our existing pages by 1.0–1.42 points.
The two judges differed by 0.87 points, more than double the 0.42 gap between the models. The results set the model we use in production.
Experience
An education crowdfunding platform, built from zero.
Payments over $1M a month across Stripe, PayPal and Razorpay; email at 500K+ a month.
Moved the whole stack to Docker and Kubernetes, cut developer onboarding from 2 days to 2 hours, and led SOC2.
Rebuilt the storefront with embeddable widgets, white-labelled domains and themes, and built Omnify Custom Payments on Stripe Connect.
I lead a team of 12 across AI engineering, AWS architecture, payments and security, for 500+ businesses.
Tech Lead for an enterprise AI agent and its AWS platform; earlier, AI/ML engineer on Storyteller.
About
I studied computer science at Anna University, and later completed an online PG Program in AI and Machine Learning from UT Austin McCombs (2023–24).
Day to day I work in Python, TypeScript and Go, on AWS and GCP, with Claude, Bedrock and OpenAI models.
I'm based in Bengaluru.
Contact
I'm always happy to talk about AI engineering, agent systems and leading engineering teams.