Hi, I'm Adhya Dagar.

Over the last three years, I've sat on both sides of the table in AI. Running data ops and research partnerships, I worked with researchers at frontier labs like Anthropic and DeepMind, building evals and data strategy that shape how models behave. Now I'm on the other side, putting agentic products in front of enterprise creative and marketing teams and getting them to work in real production.

What connects the two is evaluation. I think of evals as the product spec. Automated evals are useful at scale, but when building something new, I like spending a lot of time on the qualitative side: reading, word by word, how users actually talk to the model (rarely how we planned), and studying how the system responds. I start with whether the model actually got the job done. Then come the things beyond the outcome: tone, style, and the inputs that trip it up. That work became the foundation of how I build products.

At Adobe, I'm on Firefly's 0→1 team, building agentic systems for enterprise creative teams and taking them from first prototype to paying customers. Along the way, I'm learning how AI actually gets sold to enterprises, where adoption stalls, and what it takes to scale it into something people rely on in production.