Talvinder Singh

Talvinder Singh builds production AI systems — agents that plan and self-heal, grounded retrieval, cost-aware model routing, and the evaluation and governance that keep them honest. Serial founder (ex-OYO, YC, 500 Startups). Based in Bengaluru.
Author

B. Talvinder

Talvinder Singh

AI Systems Builder · Serial Founder · ex-OYO · YC · 500 Startups

I build production AI systems — and the companies around them.

Not demos. Systems that run in front of real users: agents that plan, retry, and self-heal; retrieval that stays grounded instead of hallucinating; model routing that picks the right model for each task and fails over when one goes down; evaluation harnesses that score every output before a human sees it; and the governance, spend controls, and audit trails that keep all of it honest at scale. Most of the industry is still shipping the demo. I work on the part that has to survive a real Tuesday.

Right now that means two things. Ostronaut — an AI-native platform that turns unstructured content into interactive learning: slides, videos, quizzes, and practice games, generated and quality-checked by a multi-agent pipeline. And an executive AI thinking partner for business and leadership decisions — one that answers first, adapts its depth to the stakes, grounds its answers in a real corpus, and self-checks before it replies. Building it in the open is also the live case study for the AI cohort I run.

That’s the current chapter of a longer arc — four companies across four technology eras in eighteen years. Digital PACE: K12 ed-tech built and sold out of college, years before ed-tech was a funded category in India. Tushky: India’s first leisure experiences marketplace (500 Startups Batch 6, Spring 2013), five years before Airbnb launched Experiences. Pragmatic Leaders (YC W21): trained 10,000+ product professionals on an ML-driven LMS before upskilling became a post-COVID gold rush. And Zopdev: cloud governance and cost automation, where an agent right-sizes infrastructure inside a customer’s own cloud. Between them: Product Head at OYO, where I led the marketplace-to-owned-inventory pivot.

I still write code. I still think in systems. I still believe the best technical decisions come from first-principles reasoning, not cargo-culting whatever’s trending on Hacker News.

This site is where I think in public. Frameworks is for thinking tools — essays where I’m working through a named idea, not just presenting a finished one. Build Logs is where I document technical decisions, architectural patterns, and lessons from actually shipping AI systems. Field Notes is Side B — founder stories, poetry, philosophy, and the thinking that doesn’t fit neatly anywhere else.

I’m based in Bengaluru. I believe in deep work, intellectual honesty, and the power of writing as a thinking tool.

What I Write About

  • Production AI — multi-agent systems, retrieval and grounding, model-routing economics, evaluation harnesses, and the gap between the demo and the system that survives real users
  • AI infrastructure & governance — why the runtime, the guardrails, and the cost controls are the product, not the model
  • India tech — why Indian SaaS has structural advantages in reliability, the evolution of the PM role, building for cost-sensitive markets
  • Leadership & product — first-principles decision-making, compensation strategy, the craft of building things that actually work

What I’m Working On Now

  • Ostronaut — AI-native learning: unstructured content in; interactive slides, videos, quizzes, and practice games out, generated and validated by a multi-agent pipeline.
  • An executive AI thinking partner — a decision-grade chat product with a real response policy: answer-first, depth-adaptive, grounded, self-checked. Also the live case study for the AI cohort I run.
  • Zopdev — Cloud governance and DevOps automation. Zopday (PaaS in your own cloud) and Zopnight (autonomous cloud-spend governance that right-sizes infrastructure inside your own cloud). The most technical thing I have ever built.
  • Pragmatic Leaders — Product and tech-leadership training. YC W21. 10,000+ professionals; 1:1 coaching is the core; an AI cohort is the newest chapter.
  • Side B — Offline-first IRL music community. The bet on real-world growth in an algorithm-saturated world. Why I started it →
  • The AI × Biology thesis — mapping the evolution of artificial intelligence against biological evolution. The first piece is live; more as the framework develops.

Get In Touch

Contact page routes by intent — coaching, Zopdev, writing, podcasts, Side B. Or jump straight to LinkedIn or X.

If something I wrote made you think differently about a problem, I’d genuinely love to hear about it.