From war survivor to AI fintech founder: Naré Vardanyan of Ntropy

Executive overview

Growing up during the Armenian war with minimal electricity, Naré Vardanyan built a career path from UN volunteer to London startup founder. After a failed first startup and a visa-driven pivot into entrepreneurship, she co-founded Ntropy, which uses language models to make messy bank transaction data readable at scale.

Horizontal platforms are hard to sell — Ntropy learned to focus on a defined customer type before going broad. Raising a seed round while pregnant forced Naré to confront conflicting identities head-on.

Adversity creates a low tolerance for settling — and that intolerance is the engine.

Path from Armenia to founding Ntropy

• Left Armenia at 17; first trip abroad at 14–15 sparked ambition to engage with the wider world
• Joined the UN's Young Professionals programme, working on financial inclusion at UNDP
• Moved to London for a master's at UCL, where proximity to DeepMind sparked an interest in AI
• Visa constraints meant the choice was return home or start a company — chose to start
• First startup, Mindbin, aimed to detect mental health issues from phone typing patterns
• Mindbin collapsed when Apple's keyboard privacy changes made data collection impossible
• Took a role at London Co-Investment Fund to pay bills while planning the next move
• Met co-founder Ilya (ex-Microsoft Research, PhD) at a Techstars demo day; complementary skills and a shared immigrant background became the foundation

How Ntropy works

• Every payment creates a text string in bank and network systems; banks with millions of customers generate enormous volumes of these strings
• Most bank systems are 30+ years old, with no standard format — producing fragmented, unreadable data
• Ntropy builds language models that interpret this data regardless of source, format, language, or currency
• Larger models generalise better, making the approach more robust than prior rule-based methods
• Cold-start problem: needed good training data to work, but needed customers to generate data
• Solved it by building a consumer page where users uploaded their own transactions in exchange for analysis — this became the first training set
• Shipped to customers early despite imperfect results; each use improved the model

Go-to-market lessons

• Building a horizontal data platform is hard to market and hard to position
• Tried to avoid niching down to preserve data diversity — this delayed growth
• Key lesson: even a horizontal product needs a defined customer type, a specific use-case language, and a focused segment before expanding
• 100 customers and 400% growth came after sharpening focus

Founding while pregnant

• Discovered pregnancy two days after seed funding hit the bank account
• Fear centred on conflicting identities: CEO, mother, woman — and the risk of resentment if one suffered
• Worried about disclosing to the board and co-founder; the reality was far more supportive than anticipated
• Core takeaway: the fear of settling — of having to stop — was the hardest part, not the logistics