Ravi Kiran Yalamanchili

Product & Engineering Leader | AI + Healthcare | 0-to-1 Builder

San Jose, CA · ravikiran2005@gmail.com · (412) 327-5858

13+ years shipping AI products in regulated healthcare. Sole PM at two companies. Took CurieAI from blank page to 10,000+ patients and ~$100K MRR.

Links

How I think about product

I’ve been sole PM at two companies and a cofounder at one. These are the principles I’ve actually operated by, not aspirational ones.

Cut ruthlessly. The best product decisions I’ve made were what I chose not to build.

At READE.ai, competitors offered 100+ EEG visualizations. Replicating that made our MVP unshippable. I ran structured discovery with neurologists and found 3 charts accounted for 80%+ of daily clinical use. We shipped those 3, cut months of build time, and delivered exactly what surgeons needed in the OR. Scope is the primary tool a PM has. Most teams underuse it.

Compliance and trust compound. Short-term margin usually doesn’t.

At CurieAI, our board pushed margins from ~15% to 30-40% through faster monetization, and a tracking feature was scoped to hit revenue targets. Compliance experts flagged it exceeded HIPAA minimums. I chose to build it fully compliant anyway. That cost us revenue in the quarter. It also produced 85% monthly retention against a 30-40% industry average, which is what made the enterprise revenue story credible in the first place.

Ship, then learn. The fastest path to a real answer is a working surface in front of a real user.

As sole PM at READE I prototyped directly in Claude Code and Cursor, pushed frontend changes to the repo, and skipped the Figma-to-handoff cycle when speed mattered. PRDs are useful. Shipped prototypes in front of clinicians are more useful. I bias toward the second.

Practitioner, not observer. I’ve built the AI workflows I’d be shipping as a PM.

At CurieAI I shipped agentic workflows with LangChain, LangGraph, and GPT-4 to turn clinical conversations into structured outputs. At READE I built real-time cerebral ischemia detection from EEG data end-to-end. I know where LLMs break down in production, which evals actually catch regressions, and when to hold the line on human-in-the-loop. That intuition matters more than ever for AI-native product decisions.