May 2026
YC held Startup School in India for the first time. Bangalore. I wanted to apply for the Bangalore one but somehow ended up applying for the SF event instead. Classic. A friend texted me asking if I'd gotten the confirmation email. I checked. Wrong event.
Applied for the right one. Got the confirmation a few days later. Booked flight tickets that same day. Reached Bangalore running on caffeine and the vague hope that this wasn't a waste of a flight.
It wasn't.
One day of founders talking about the parts of building companies that nobody writes blog posts about. Not the "how we got to product-market fit" victory laps. The messy middle. The parts where you're not sure if what you're building matters, and the only signal is that users keep coming back (or don't).
I went in as someone who builds things. I came out thinking differently about why I build things. There's a gap between "I can ship this" and "someone needs this shipped." Most of my projects live on the first side. The best ones would live on both.
No sleep. A lot of coffee. Worth it.
The day after Startup School came ContextCon, a hackathon run by YC and Crustdata (YC F24). Five hours to build something. My teammate Soham and I built an AI-native recruiting platform.
The problem: recruiters spend 70% of their time on work that produces zero signal. Searching the same role under different titles, enriching profiles one by one, writing custom outreach messages. It takes weeks when the actual evaluation takes minutes.
What we built: a single-prompt system. Type what you're looking for. It finds candidates, scores them, and drafts personalized outreach. Days of work compressed into minutes. Built on Crustdata's infrastructure for the enrichment layer.
We finished 4th out of all the teams. Five hours from nothing to a working demo. Here's the LinkedIn post with more details.
The hackathon validated something I'd been circling around: the gap between "AI that works in a demo" and "AI that works for a real workflow" is almost entirely about understanding the workflow. We didn't win because of better models. We placed because we understood what recruiters actually spend their time doing.
Same lesson as DevCard, same lesson as GSoC. The engineering is the easy part. Understanding the problem is the hard part. Everyone says this. I'm starting to believe it.