Sankalp Gulati

Sankalp Gulati

Chief Data Scientist · EkaCare

I work across domains that have little obvious in common — Indian classical music, private capital markets, clinical AI — but they keep asking the same question: how do you build intelligence for a world that doesn't arrive pre-labeled, pre-structured, or in English?

Right now, I'm building AI for Indian healthcare. The scale is enormous — over a billion people, most of them accessing care through a system that runs on paper, on memory, on overworked doctors. The diversity is staggering: twenty-two languages, scripts, clinical vocabularies, accents, handwriting styles. General-purpose AI doesn't cut it here. What's needed are models that actually know this specific context, can be trusted in a clinical setting, and work at Indian scale. That's what I'm trying to build.

Before that, I spent a few years in the world of private capital — using machine learning to read the economy through unconventional signals. What do employee reviews tell you about a company's trajectory six months out? Can you map competitive dynamics across thousands of private companies from how they talk about themselves online? That chapter was about one thing: finding truth in data that wasn't designed to contain it.

It started with music. I spent a decade trying to teach computers to understand Indian classical music — not just recognize it, but grasp something of its internal logic. A raga isn't a scale; it's a living melodic grammar that unfolds differently every time. Formalizing that computationally, from raw audio, is still one of the most beautiful problems I've encountered. It taught me that the most interesting AI challenges are the ones where human knowledge is deep, culturally specific, and hard-won.

The thread running through all of it: AI is most valuable when it closes the gap between what a domain expert knows and what everyone else can access — the classical guru's ear, the seasoned analyst's intuition, the well-trained doctor's attention — made available to anyone, anywhere.

When not doing any of this, I'm probably dialing in a pour-over, badly noodling on a guitar, or convincing myself that this next photo will finally nail the shot.

PhD, Music Technology Music Technology Group, Universitat Pompeu Fabra · Barcelona
2016
Master's, Sound and Music Computing Music Technology Group, Universitat Pompeu Fabra · Barcelona
2012
B.Tech, Electrical Engineering Indian Institute of Technology (IIT) Kanpur
2008
International Society for Music Information Retrieval (ISMIR) 2013–2018 · 5 years
IEEE/ACM Transactions on Audio, Speech, and Language Processing 2016, 2017
IEEE Transactions on Multimedia 2017
Journal of New Music Research (JNMR) 2016
EURASIP Journal on Audio, Speech, and Music Processing 2018
IEEE Journal of Selected Topics in Signal Processing 2018
Journal of Intelligent Information Systems (JIIS) 2016
Speech Music and Mind (SMM) 2018
National Conference on Communications (NCC) 2015
Audio Signal Processing for Music Applications (ASPMA) Coursera MOOC · Music Technology Group, UPF · 2014–2015 Designed and implemented programming assignments and automated assessment tools for a global online course.
Processament de So i Música Universitat Pompeu Fabra, Barcelona · 2013–2015 Teaching assistant — conducted seminars and lab sessions, created programming assignments with automatic graders.