I was in Delhi last week for the AI Bharat Summit. I hadn't planned to make the IIT Delhi visit part of the same trip, but it ended up fitting — and the two days together left me with a lot to think about.
AI Bharat Summit: A Snapshot of Where the Ecosystem Is
There's something clarifying about being in a room full of people who are genuinely building — not talking about building, not advising on building, but actually shipping. The Bharat Summit had that energy.
I spent time with Ayush, who heads Voice AI at Plivo, and crossed paths with Vivek Raghavan from Sarvam. A lot of the conversations were about enterprise AI — what's working in production, what's harder than expected, where the next set of problems is coming from.
What I found interesting was less what people said in the sessions and more what kept surfacing in the quieter moments between them. A kind of shared intuition that the infrastructure layer is largely solved — models are good, pipelines exist, integrations are possible — and that the harder question is somewhere above that. Not can the AI respond well? but can the system reliably get to the right place, every time, in a way you can actually stand behind?
Nobody named it the same way twice. But it was the same question.
(Side note: I took the Delhi Metro between venues. On time, clean, no traffic. As someone who lives in Bangalore, I'm still recovering from the shock. Also, I have exactly one photo from the entire trip — next time I'm treating it like a proper travel blog and documenting everything.)
IIT Delhi: When Researchers Ask the Hard Version
From the summit, I went to IIT Delhi. Naveen, who heads the CSE department, was kind enough to connect me with two professors working in systems research. The kind of meeting where you walk in with a deck and walk out having rewritten half of it in your head.
Researchers don't ask about market size. They ask about edge cases, failure modes, what happens when your assumptions are wrong. Presenting what we're building at Formi in that context — an architecture that treats uncertainty as a first-class system, that enforces boundaries deterministically, that separates what it knows from what it's inferring — forced a level of precision that I think made the work better.
The collaboration we're exploring with IIT Delhi is partly about that. Bringing genuine systems thinking to bear on a problem that deserves it.
What I'm Taking Away
I don't think either day gave me answers so much as better questions.
The summit showed me a market that's maturing faster than most people expected, full of smart teams solving real problems. The IIT Delhi conversations reminded me that the hardest engineering problems don't announce themselves loudly — they show up quietly in the edge cases, in the moments when a system that looked robust turns out not to be.
At Formi, we're working on something that we think sits at the intersection of those two observations. What it means to go from a great AI interaction to a guaranteed, auditable, enterprise-grade outcome — at scale, under real conditions, in the kinds of environments where the stakes are high and the margin for error is low.
We're getting close. More soon.