5 Questions with Romain Clément, Founder & CEO of ArcaScience

In a recent keynote this year, you shared your personal journey as a patient for the first time. Can you tell us how that defining moment led to the founding of ArcaScience?

I was 19 when a radiologist pointed at a 9-centimetre mass on my MRI scan. I was sitting there with my mother, and the diagnosis was handed to me as a cold, clinical statistic. The medical system didn’t have a full, integrated picture of me; it only had the statistical average of what that shadow usually meant for the median patient.

At the time, I was just a university student coding in my free time. That moment made me realise a fundamental flaw in modern medicine: patients are treated as averages. The system completely lacks the infrastructure to prospectively model how a specific treatment will actually interact with an individual’s life, balancing its unique benefits and risks.

The technical blueprint for ArcaScience was born right there. Our entrepreneurial goal became “Total Integration.” We set out to build an AI infrastructure that treats every piece of fragmented data – clinical papers, safety reports, real-world data points, and unstructured patient journeys – as a connected thread in a single graph. We engineered our architecture to eliminate generic, “standard” protocols by providing a continuous, auditable flow of truth. I wanted to build the technology I needed to see on that screen when I was 19: clarity, personalisation, and a clear path back to health.

From a product and market perspective, why is relying on a “Standard Protocol” so dangerous in modern medicine and drug development?

In tech terms, a “standard protocol” is the ultimate product compromise. It’s a solution designed for a “median user” who doesn’t actually exist.

When drug developers and clinicians rely on these broad standards, they are essentially admitting that their data infrastructure isn’t powerful enough to specialise. This is a massive market failure. Right now, the vast majority of drugs in the R&D pipeline fail because companies are trying to build “one-size-fits-all” software for the human body.

The danger lies entirely in data opacity. If an AI or a researcher cannot pinpoint exactly why a molecule helps “Group A” but harms “Group B” long before a trial even starts, they aren’t practising data science; they’re playing a very expensive game of probability. At ArcaScience, our product mission is to replace data opacity with what we call Precision Reality.

In 2020, ArcaScience was selected by the French government to help tackle the COVID-19 crisis. As an entrepreneur, how did that high-stakes environment force your platform to evolve?

When the pandemic hit, governments didn’t need a basic search engine – they needed a predictive decision engine. There was a mountain of fragmented, chaotic data: unreviewed pre-prints, early global trial results, and scattered molecular signals. Our platform was tasked with harvesting, cleaning, and merging these disparate datasets in real-time to identify existing drugs that could be immediately repurposed to save lives.

This was a massive, high-stakes stress test for our technical infrastructure. It served as our ultimate proof of concept for speed and data scale. It proved that our AI could harmonise global data chaos into actionable insights under extreme pressure. More importantly, it forced us to refine ArcaScience’s core product focus around “Prospective Benefit-Risk Evaluation.” It taught us that our platform couldn’t just surface “potential hits” – it had to provide the deep transparency and algorithmic rigour required for enterprise-grade, life-or-death decision-making.

You’ve spoken about a turning point involving a cancer patient who reached out to you early on. How did that encounter directly reshape the actual code and architecture of your platform?

In 2020, a patient with a severe brain tumour reached out to us for help, and at that stage in our development, our platform simply wasn’t ready to support her. Going home in silence that night was a defining moment for me as a tech founder.

At that time, our software was excellent at finding data signals and generating hypotheses, but identifying a lead is entirely different from backing a critical clinical choice. To truly move the needle in highly complex sectors like oncology, we realised we didn’t need a tool that simply ranked options. We needed an unshakeable, auditable model that provided an explicit “Confidence Score.”

We realised our product had to change. We went back to our foundational infrastructure and engineered a proprietary stack of 24 specialised AI models. To completely eliminate the “hallucination” issues plaguing mainstream Generative AI, we hard-coded a strict Traceability Engine. This allows users to “walk back” every single insight generated by the AI to its exact, verified source document.

We even brought on one of the original computer scientists behind the Transformer architecture (the underlying tech behind ChatGPT and Claude) to ensure our data models were grounded in absolute rigour. We shifted our entire codebase from asking “What does the data say?” to “What is the most auditable, risk-mitigated path for this specific profile?” That patient became our North Star.

Despite massive technological advances, 95% of drugs still fail during clinical trials – a staggering stat for any industry. What does the market look like when ArcaScience’s “World Model” becomes the default standard?

It looks like a world where we finally stop gambling billions of dollars on volatile molecules.

Think about it: as a society, we can land probes on moving comets because physics is entirely predictable. Drug development only feels unpredictable because our data is siloed and our AI models are poorly framed. By integrating all qualitative and quantitative evidence into a single coherent, auditable World Model, we turn a guessing game into predictable engineering.

When this architecture becomes the industry standard, that 95% failure rate becomes a historical relic. Instead of running a data post-mortem after an expensive Phase III trial fails, companies will use prospective modelling long before Phase II. It means life-saving solutions will hit the market years faster, and data-driven personalisation will replace broad compromises. For founders and investors, it means shifting biotechnology from a high-risk lottery to a scalable, predictable tech stack.