The co-founder who taught software to read a feeling

There is a photograph of the three people who built Entropik. Two of them stand where your eye lands first. The third, in a plain dark suit, holds the right edge of the frame, arms folded, half a smile. That is Bharat Singh Shekhawat, and the position is fitting. He is the co-founder who runs the part of the company you never see - the engine that decides whether the whole idea actually works.

The idea is a strange one to build. Entropik measures emotion. Not in the loose way a survey does, asking people to rate a feeling on a scale of one to five, but by reading the body directly. A face tightens. A voice climbs. The eyes drift to the corner of a screen and stay there a beat too long. Each of those is a signal, and Shekhawat's work is to catch them, clean them, and turn them into something a brand can trust enough to make a decision on.

He has been doing it since 2016, when he and his co-founders started the company in Bengaluru. Nine years later he still carries the title Head of Engineering, which tells you where his attention has stayed. Plenty of technical founders drift toward the stage over time. He drifted toward the codebase.

01 / The problemMeasuring the thing you were told you couldn't

For most of its history, market research asked people what they thought and wrote down the answer. The trouble is that people are unreliable narrators of their own reactions. They forget. They flatter. They tell you what they think you want to hear. A gap opens between what someone says about an ad and what their face did in the two seconds they watched it.

Entropik was built to close that gap, and the founding bet was blunt: emotion leaves a physical trace, and a trace can be measured. The company's name plays on entropy, the measure of disorder in a system. It is a good name for a business that takes the messiest signal in the world - a human reaction - and tries to make it legible.

The commercial logic follows quickly once you accept the premise. Every ad, every product screen, every checkout flow produces a reaction whether or not anyone measures it. Brands spend enormous sums on the assumption that they can guess which of those reactions is positive. A tool that reads the reaction directly, rather than asking about it afterward, changes the economics of that guessing. It is why an engineering-heavy company out of Bengaluru ended up with customers on five continents rather than a science paper and a shelf.

"We're moving from being analytical and prescriptive to being predictive."

That sentence is a mission statement and an engineering roadmap at once. Describing what already happened is hard. Predicting what a person will do next, from the same handful of signals, is much harder. It is the difference between a report and a forecast, and closing it falls squarely on the technical side of the house.

02 / The stackFour channels, one read

The platform Shekhawat helped architect does not lean on a single tell. It reads four channels at once and weighs them against each other, which is what "multimodal" means in practice. A smile alone is easy to fake. A smile that disagrees with the voice, or with where the eyes went, is more honest.

Running four channels in real time, for enterprise customers on five continents, is not a science-fair demo. It is a distributed systems problem. The data is heavy, the latency budget is tight, and the output has to be steady enough that a client will bet a product launch on it. This is the quiet discipline underneath the flashy label - reliability, at volume, on a signal that was never meant to be measured.

Weighing the channels against each other is where the engineering earns its keep. Any one signal can mislead. A face can be polite. A voice can be performing. Gaze can wander for reasons that have nothing to do with interest. The read only becomes trustworthy when the system knows how much to lean on each channel in a given moment, and when to discount one because the others disagree. That balancing act is not a slogan on a landing page. It is model design, data pipelines, and a great deal of testing that no customer ever sees.

03 / The pathRajasthan, C-DAC, and a long apprenticeship in systems

Shekhawat came to the problem the unglamorous way. He studied Computer Science and Engineering at the University of Rajasthan, then took advanced technical training at the Centre for Development of Advanced Computing. Before Entropik he had already spent years close to production systems, working as a software engineer building data-heavy platforms and later serving as Vice President of Technology at Redcastle. He also founded an earlier venture, Bigpaa, of his own.

By the time Entropik started, he had more than a decade of building behind him. That matters. Emotion AI sounds like a research toy, but a company selling it to Fortune 500 buyers needs the boring virtues - uptime, throughput, a system that behaves the same on a Tuesday as it did on the day of the pitch. Those are earned over years of shipping, not invented in a hackathon.

The earlier roles also explain the temperament. Running technology at another company, and building a startup of his own before Entropik, is the kind of experience that teaches a founder where systems tend to break and how much of the work sits in the parts nobody demos. Someone who has already carried a platform in production tends to respect the difference between a feature that impresses in a meeting and one that survives a year of real traffic. That respect is exactly what an insights company needs at the top of its engineering.

Emotion is supposed to be the thing you cannot quantify. A small team in Bengaluru disagreed - and someone had to make the disagreement run in production.

04 / The buildNine years, one throughline

Trace the company's product history and you can read the engineering ambition growing underneath it. Affect Lab arrived in 2018 as the first multimodal Emotion AI platform. Voice capabilities were folded into the stack over 2021 and 2022. By 2024 the team had shipped Qatalyst, a UX research platform, and the Mira AI Moderator, which climbed to number five Product of the Day on Product Hunt.

The funding tells its own version of the story. Entropik started on a seed round of $150,000 in 2016 - a genuinely small number for a company attempting something this technically heavy. The $25.5 million Series B in 2023 came seven years and a great deal of building later. In between sits the unglamorous middle of a startup, where the work is mostly making the thing more reliable than it was last quarter.

05 / The roleThe third founder problem

Every founding team eventually splits the work. Someone carries the market. Someone shapes the product. Someone has to make sure the machine underneath does not fall over. In Entropik's telling, the market and the message have public faces. Shekhawat took the third seat, the engineering one, and stayed in it.

It is the least visible of the three jobs and, in a company like this, arguably the one with the least room for error. A pitch can be re-recorded. A product can be re-designed. A platform that quietly returns the wrong read on 45,000 tested experiences is a different kind of problem - the kind that erodes the one thing an insights company sells, which is trust in the number.

The best compliment an engineer can earn is that the product just works. No one thinks about the architecture until it breaks.

There is something clarifying about that constraint. Shekhawat's name does not headline the interviews, and his surname - Shekhawat, a Rajput clan name out of the Shekhawati region of Rajasthan - carries more history than his public profile does. He seems content with the arrangement. The proof of his work is not a quote. It is that the platform runs, at scale, on a signal the industry spent decades insisting could not be measured.

06 / What's nextFrom what happened to what will

The forward bet is prediction. Telling a brand how people reacted to an ad is useful. Telling them how people will react, before the money is spent, is the product Entropik is reaching for. That is the hardest version of the problem, and it lands where it always lands - on the engineering. Better models, cleaner signals, a platform steady enough to stake a forecast on.

For a co-founder who has spent nine years making messy human data behave, it is a natural next chapter. The stage-facing work at Entropik will keep drawing the attention. The engine will keep needing someone who trusts the data more than the demo - and, for now, that is still the same person who took the right edge of the frame and folded his arms.

If there is a lesson in his career, it is a quiet one about where value hides in a company. The vision gets the headline and deserves to. But a vision about measuring emotion is only as good as the platform that measures it, and platforms are built by people who are willing to spend years on the unglamorous middle. Bharat Singh Shekhawat has been in that middle since 2016, and Entropik's numbers - the customers, the patents, the tested experiences - are the receipts for the time he put in.