All Posts

I’ve realized something about my own behavior recently:I’ve almost completely shifted from search engines to conversational AI. From random questions to specific tasks, the ease of getting answers now feels fundamental. And the moment I leave AI and enter traditional apps, the friction becomes obvious. Two types of digital interactions If you zoom out, most of what we do online falls into two categories: 1. Exploratory - Shopping, travel, content discovery → Users are still d

I built an AI-powered notes app to fix what Apple Notes doesn’t do well. It’s great for storing notes—but not for organizing or understanding them. So I built one that: Auto-organizes notes by intent, not folders Cleans up messy thoughts automatically Lets you search by meaning, not keywords From passive storage → to an active thinking assistant. Key Insight: Raw AI is powerful. Production AI needs scaffolding. I went in thinking: “Give it a prompt → get structured JSON → shi

Lessons from building with prompts, agents, and code—and the one question that matters most. 🧠 The Missing Half Most of us aren’t great prompt writers. When we ask a question, we often communicate only half of what’s in our head. The context, the goal, the constraints—much of it stays unspoken. The AI systems that feel magical are the ones that infer the missing half. ✈️ Ask about a long-distance trip → it accounts for jet lag, fatigue, and pacing instead of just dumping a g

I was one of the founding members building user-centric evaluation frameworks at Meta AI. When I entered the GenAI space, most evaluation approaches were rooted in research benchmarks—valuable, but often disconnected from how users actually experience AI in the real As an early member working on text response quality at Meta, I saw this gap clearly. So I approached evals as a product problem, not just a measurement problem. Evals are the control system for GenAI products. The