What I believe

I believe startups are how new solutions reach the world. Most of the things that have materially improved human life in my lifetime — the way we communicate, the way we move, the way we learn — came from small teams who set out to solve a real problem and built the engineering to back it up.

Engineering is how ideas become real. Without it, a startup is a pitch deck. With it, a startup becomes a company that changes something.

But engineering at a startup is not engineering at Google. The decisions are harder, the information is thinner, and the stakes are higher. A senior engineer at a Series A company is making choices every day that would have a team of architects at a big company. They usually make them alone, with Stack Overflow and a gut feeling.

I want to change that. My work is to take the judgment I've built over 18 years — and the pattern recognition I'm still building as a practicing architect — and make it available to the people who need it. Not as a course. Not as a book. As evidence-based evaluations and decision frameworks that are free to read, rigorous enough to trust, and specific enough to act on.

What I love most in this work is discovery — the moment when an experiment reveals something nobody had noticed before. My job, as I see it, is to do that kind of work on the tools startup teams need to choose between, and to publish what I find so they don't have to run every experiment themselves.

What I'm doing now

Right now, the highest-leverage thing I can give startup teams is clarity about agent engineering. Claude Code, Cursor, GitHub Copilot, and the whole landscape of agent skills and plugins are reshaping how software gets written. Most of what's being published about them is either marketing or thin opinion. There's very little rigorous evaluation, and almost none of it is written from a startup perspective.

So I'm building startupengineering.io as an open, evidence-based evaluation of the agent engineering tools that startup teams need to choose between. Every skill I evaluate is tested against the agentskills.io best-practice rubric and run through real scenarios on multiple agents. I publish the methodology, the data, and the decision frameworks openly.

When the field shifts — and it will — the focus will shift with it. Startups won't stop needing better engineering decisions. The tools they need to decide about will just be different.

My role in the ecosystem is the translation layer. I take what the frontier labs and skill authors produce, test it in real startup contexts, and publish the evidence. I don't invent the new tools — that's the researchers' job. I don't execute production systems for clients — that's the builders' job. I work in the gap between them, where the question “does this actually help in practice?” gets answered.

Where I've been

• 2021 – present · Principal Architect at Mintmesh India · Leading a 14-person engineering team building Gen AI SaaS. Transitioned legacy AI to Gen AI in 8 months. Improved accuracy from 60% to 85%. Built LLM orchestrator at 50M TPM. Reduced OpenAI costs by 5x. Designed the AI observability stack.
• 2016 – 2021 · Founder & CEO at Naanal Technologies · Built a logistics SaaS from 0 to 45 people and 6 crore INR ARR. Led a 30-engineer team. Shipped a rule-based AI decision engine handling 50+ parameters that improved on-time delivery by 30%.
• 2010 – 2016 · Staff Software Engineer at Brocade Communications · Architected software load balancers using Intel DPDK that hit 160 Gbps throughput. Demonstrated the partnership work at VMworld 2012.
• 2005 – 2008 · Member Technical Staff at HCL Technologies · Built network security modules for Juniper's mobile backhaul router.
• M.S. Computer Engineering, NC State · 2008–2009
• B.E. ECE, Government College of Technology Coimbatore · 2001–2005