After three days of using Vivid, it was perfectly clear to us: Vivid is the right partner for Leap Dynamics. And we have stayed ever since.
Magdalena Genov — Co-Founder & CEO, Leap Dynamics
On a desk in a Berlin office sits what most people would consider a throwaway machine: an old laptop with 16 gigabytes of RAM and no dedicated graphics card. On it, an 8-billion-parameter AI model runs at roughly 30 tokens per second. “Without a GPU, that is impressive,” says Adrian, AI developer at Leap Dynamics GmbH. The laptop is not a relic. It is a proof of concept — and proof of concept is precisely the business Leap Dynamics is in.
The company was founded in 2019 by Nikolai and Magdalena Genov, two bioinformaticians who had spent years in academic research before concluding that what they knew how to do — build custom software, analyse large and opaque datasets, make sense of biological and medical data — was badly needed in the commercial world. Today the Berlin company serves clients ranging from large banks and pharmaceutical firms to small businesses cautiously navigating a wave of AI hype. The question Leap Dynamics keeps asking is not “can we use AI here?” but “should we?”
From Genome to Market
Nikolai Genov holds a PhD in bioinformatics. During his doctoral research he worked extensively with cancer gene-expression data, spending as much time in front of a terminal building custom analysis pipelines as he did in any laboratory. By 2016 he had begun working seriously with machine learning, well before it became a business buzzword. The bioinformaticians had certainly recognised, he says, “that these systems were very good at classifying biological data, even then. The potential was there.” At a systemic level, though, it was not acted on. “I believe it was simply not seen,” he adds.
“I had the feeling,” Nikolai explains, “that if we simply did what we already knew how to do — but more organised, outside the academic environment — we would be successful. So far that has proven true.”
Magdalena brought a complementary background: after completing her bioinformatics degree, she spent several years in space medicine research at a major research hospital, where she explored extreme environments and their effects on the human body — from experiments aboard the International Space Station to studying the physiological impact of climate change. Her responsibilities included organising parabolic flights and preparing scientific experiments for the ISS. “In a very short time you have to take care of a great many things at once — ideally everything should have been done yesterday.”
“Being a researcher means dealing with questions for which there is not yet a right answer,” Magdalena says. “And when you are an entrepreneur you very often deal with clients who produce large volumes of data but also do not know what to do with that data.” How can those data be applied so as to generate new clients, new products, more liquidity? The question is the same one. “In the end you transform yourself, as an entrepreneur, into a kind of researcher.”
The Golden Hammer Problem
In the current moment Leap Dynamics occupies an unusual position: a company that regularly tells clients the technology they came asking for may not be what they need. “Golden hammers,” Adrian calls them — an Anti-Pattern, a tool someone applies to every problem regardless of whether it fits. “What you very often see is that AI is thrown at every problem. That is highly wrong. There are problems where classical signal processing or classical image processing is far, far more appropriate.”
The methodology Leap Dynamics uses to navigate this is the Proof of Concept: a rapid, contained test of whether an approach works under real conditions, before any project escalates into a full development cycle. When a question arose about whether AI could run without expensive graphics hardware, the team simply ran it on an old laptop and measured. The model performed at an acceptable speed. Proof established. “If this Proof of Concept is successful, you can apply best practices and expand the whole thing so that it could go towards production.”
A more commercially consequential example concerns data privacy. A client needed personally identifiable information — names, addresses, telephone numbers, bank details — automatically anonymised from documents, without routing them through a cloud service. “We had the Proof of Concept: can you do this without major hardware, but with modern AI? We did it, and we have taken that into a product we are currently bringing to market.”
For clients arriving with larger ambitions, Nikolai is candid about the sequencing. “In many companies, digitalisation has not progressed as far as one would like. That means digitalisation steps still have to happen before an AI system makes any sense at all.”
For Adrian, the through-line across every project is a single principle. “AI is an amplifier,” he says. “It amplifies what is already there. It worsens what is already there when bad things are present — and it improves what is already there when good things are present.”
Data That Stays in the Building
When the team debates which cloud AI providers to recommend, Adrian’s position for clients with sensitive data is consistent: neither of the large American ones. “It would take only a signature from the US president and German business secrets are suddenly public,” he says. “That risk is simply not present with a German AI or a German server.”
This is not a theoretical concern. Nikolai describes how the company routinely works with medical data, genetic data, and financial data — precisely the categories that cannot legally or safely be routed into a third-party cloud model. “When we use AI, the question is always: how do we make that use safe — in accordance with the DSGVO, in accordance with the AI Act.”
European alternatives exist and are better than their public profile suggests. Mistral, based in Paris, produces models compact enough to run on a standard desktop while handling image and text classification. Black Forest Labs, a Freiburg company, is what Germany calls a Hidden Champion: barely known outside the field, yet its image-processing algorithms are used by some of the largest technology firms in the world.
Two Founders, One Whiteboard
At Leap Dynamics the division of responsibilities is organised around strength rather than seniority. Nikolai leads the technical and data work, including all development in R, which he has practised for more than ten years. Magdalena manages the organisational and communicative side of the business: client relationships, project management, coordination, and financial planning — and of course what she openly describes as the paperwork that feels like it takes up half of every founder’s day. “Managing a company means simultaneously being a researcher, a developer, and a person who sits at a desk and has to wade through documents for half a day.”
The COVID years forced a strategic turn. Leap Dynamics had started with a hardware focus. When the pandemic arrived, the team pivoted to personalised software solutions. “That was the right decision,” Magdalena says. “We got through the COVID period very well, gained many new clients, and could then turn our attention to integrating AI into our offering.”
Disagreements go to the whiteboard. “We list all the advantages and disadvantages of a decision,” Magdalena explains. “We don’t look only at our personal preferences — at the end, we look at what will move the company forward.” The deeper principle is Nikolai’s: “It is often not about making the right decision, but simply being capable of making any decision at all. What actually destroys companies is not only that wrong decisions are made, but that decisions are postponed for five years. That is far worse.”
Adrian, who came to Leap Dynamics from a hybrid software-development and data-science background in the banking sector, distils his experience into a single insight: “One of the most important things I carry from the banking world is what becomes possible when you have a team that actually works.”
Liquid Enough to Grow
Leap Dynamics came to Vivid through an acquisition. The founders had previously held an account with Pile for business interest savings — a core financial tool for a firm that works in project cycles, with significant inflows arriving in concentrated bursts. When Pile was acquired by Vivid, the founders took the transition as an opportunity rather than an obligation. “We had the chance at the time to get to know Vivid and still choose against it,” Magdalena recalls. “After three days of using Vivid, it was perfectly clear to us: Vivid is the right partner for Leap Dynamics. And we have stayed ever since.”
For Nikolai, the original draw remains the most important one. The Interest Account turns dormant project income into working capital between engagements.
For a small company that does generate revenue and has money sitting in the account at various points — it is actually simply useful. It simply improves liquidity.
Nikolai Genov — Co-Founder & CEO, Leap Dynamics
Magdalena describes how each project can carry its own sub-account and its own IBAN, with a defined budget assigned from the outset. Team cards carry individual limits. “The fact that I can actually assign a product an IBAN with a set amount of money — it makes planning incredibly easier.” The Business Investment Account, which Nikolai was already searching for elsewhere when Vivid added it, completes the picture.
Having someone there — even if it is an AI agent — who can answer our questions or simply clear away certain uncertainties, is worth its weight in gold.
Magdalena Genov — Co-Founder & CEO, Leap Dynamics
“Using Vivid’s AI agents is the next natural step for us as Vivid customers,” Magdalena says. After seven years of building, and more than eight of thinking about what such a company could become, Nikolai is unambiguous. “I don’t regret it. Even when it is hard. In reality it is very, very hard. But it is worth it.”
