TuhinChakraborty

Mimica, the company he co-founded and runs as CEO, does something deceptively plain. It watches. As employees move across screens and systems, the software captures every click and keystroke, strips out anything sensitive, and reassembles the mess into a clean map of how the work actually gets done. Not how a policy document says it should. How it does.

That distinction is the whole business. For years, enterprises bought automation the way they buy gym memberships - with optimism and a vague plan. Someone would guess which tasks were worth automating, hire developers to hand-code each step, and discover months later that the process had quietly changed. Chakraborty's pitch flips it: stop guessing, stop programming every action, and let the software learn the job by observation first.

In 2025 that idea got a fresh and very current name: training data for AI agents. Mimica raised a $26.2 million Series B in September, led by Paladin Capital Group, on the argument that the agents everyone is racing to deploy are useless until they understand the work in front of them. “Capability means nothing without context,” Chakraborty said when the round closed. “The agents that will win are the ones that understand the work.”

He did not arrive at this from a business-school case study. He arrived as an engineer. He grew up in the Bay Area, immersed in technology, and spent the first chapter of his career inside two companies that ran on it. At Pandora he was a software engineer and eventually a senior engineer and technical lead, the kind of role where you learn how large systems actually behave under load and how often the documentation lies. A short stint as a senior software engineer at LinkedIn followed.

Then, in 2017, he left to start something. He paired with Raphael Holca-Lamarre, a co-founder whose credentials read like the other half of the puzzle: a PhD in neuroscience and machine learning. One builds systems, the other studies how intelligence learns from observation. The company they founded would do exactly that - learn from observation.

The early years were a quiet grind of getting the science to work. Backed by Entrepreneurs First and seed money from Episode 1 Ventures in 2019, the team shipped Mapper in 2020, which turned raw click-and-keystroke data into readable process maps. The validation came fast and unusually clean: after the 2021 Series A from Khosla Ventures, the company reported converting 100% of its pilots into paying customers. In a market full of pilots that go nowhere, that number is a statement.

By 2022 came Miner, the AI-powered task mining engine, and with it recognition - Everest Group named Mimica a Leader and Star Performer in task mining across 2023 and 2024. Underneath the analyst language was a simpler scoreboard: more than 30 large enterprises, multiple Fortune 500 names, and over a million hours of work handed back to the people who used to do it by hand.

Chakraborty has been careful about what kind of company this is. Software that records everything an employee does is, in the wrong framing, a surveillance product. He drew the line early and keeps repeating it: Mimica is a process improvement tool, not a performance measurement tool. The platform anonymizes aggressively - names, addresses, phone numbers, credit card details, social security numbers all stripped before anyone sees a report. The point is to fix the work, not to grade the worker.

His read on generative AI is less about hype and more about scope. The categories of work that resisted automation - communication, unstructured data, judgment calls - are suddenly in play. “Generative AI is set to dramatically increase the volume of work that can be automated,” he has said. Which is precisely why knowing what the work is, in granular detail, becomes the scarce resource. You cannot automate what you cannot see.

There is a tidy symmetry to where he sits now. He spent years inside big technology companies learning how real systems diverge from their diagrams. He now sells the instrument that closes that gap for everyone else - and, increasingly, the playbook that lets an AI agent step into a human's seat without falling over. The mission he states for it is almost old-fashioned next to the machinery: give enterprises and the people in them back their most precious resource, which is time.

Mimica is headquartered around London with Chakraborty based in New York, a transatlantic setup that suits a company selling to global enterprises in healthcare, logistics, financial services, and manufacturing. The Series B will go toward turning captured workflows into the training substrate for agentic AI - the connective tissue, as he frames it, between how work is done today and how it could be done tomorrow.

Ask him where it ends and the answer is unglamorous on purpose: AI moving from an interesting internal experiment to something that quietly runs at the heart of the business, absorbing the repetitive and the mundane so people can spend their hours on work that deserves them. He is not promising to replace anyone. He is promising to read the room - every click of it - and tell you, with data, what was never worth a human's attention in the first place.