About AlkhemYst-Ai.cloud
Research teams do not have a data problem. They have a problem with everything that happens between collecting the data and being able to defend the result. Steps get lost. A pipeline that ran on one machine will not run on the next one. The postdoc who wrote the analysis finishes their contract, and the method leaves with them. A reviewer asks how a figure was produced, and the honest answer is that nobody wrote it down, because there was never a place to write it down that was easier than not writing it down.
This platform did not begin as a product idea. It began as the same problem turning up again and again, in very different places. In doctoral research on big data for industrial systems, the finding that kept repeating was not about algorithms. It was that the science usually worked and the infrastructure around it did not. On the industrial side, the same gap showed up with higher stakes: a model that cannot be rebuilt is a model nobody will run a production line on.
Two years went into what came next. Not two years of writing code. The first part was analysis, sitting with the actual workflow of research groups and working out where the time really goes, which is almost never where people assume. Then strategy, deciding what this had to be and, harder, what it had to refuse to be. The building came after that, and it took the shape it did because the two years before it had already answered most of the questions.
Our founder has spent his career on both sides of that gap. Academic training and doctoral research on one side, and on the other, years leading field teams in demanding physical operations and putting machine learning into systems that had to keep working when nobody was watching. Those two worlds usually talk past each other. Research asks whether a result is true. Industry asks whether it holds up on a Tuesday when something breaks. A platform for research data has to answer both, and most tools are built by people who have only ever had to answer one.
So this is not a general data science tool pointed at universities. Every stage of the research process is tracked, versioned, and reproducible by default, as a property of how the platform is built rather than a feature switched on later. Data stays in Europe, on European infrastructure, because for most of the institutions we work with that is not a preference, it is the condition for using anything at all.
And because research is done by teams rather than individuals, the platform is organised the way an institution already is. People join under their real academic roles, principal investigator, lab manager, postdoc, PhD candidate, with a supervisor approving access before anyone touches data. Work is grouped into projects, and a supervisor can follow any of them from one place: progress, pipelines, to-dos, review feedback, and the final report. The aim is simple. Less time lost between people, and less work that quietly cannot be repeated.
Why this is worth building
This is not a niche complaint. When Nature asked 1,576 researchers about it, the answers were not close:
The second number is the one that matters for a platform. Most researchers have been on the losing end of this, and almost none of them ever hear about it when it happens to their own work. The failure is silent, which is why it persists.
It is not for want of trying. In ecology, one audit screened 346 papers and found that only 27% came with the code behind them. In the social sciences, a larger audit traced what happens between publication and somebody rerunning the analysis, and it is worth looking at where the losses happen:
Read the middle bar first. Two thirds of the work never ships the code at all, and that is the part no reviewer can catch. Then read the third: even among the authors who did share, half the results did not come back the same. Nobody in that study was careless. They were working the way research is normally worked, with the analysis in one place, the data in another, and the record of what was actually run living in somebody’s memory until it did not.
That is the gap this platform is built to close, and it is why reproducibility here is a property of the system rather than a discipline asked of the user. If the record has to be written by hand after the fact, it will not be written.
Sources. Baker, M. (2016). 1,500 scientists lift the lid on reproducibility. Nature 533, 452–454. doi:10.1038/533452a · Culina, A., van den Berg, I., Evans, S. & Sánchez-Tójar, A. (2020). Low availability of code in ecology: a call for urgent action. PLoS Biology 18(7), e3000763. doi:10.1371/journal.pbio.3000763 · Krähmer, D., Schächtele, L. & Auspurg, K. (2026). Code sharing and reproducibility in survey-based social research. Royal Society Open Science 13. doi:10.1098/rsos.251997
What the platform does
Open any line to see how it works, step by step.
The analysis families it covers
330 components in 15 families, each one a step you place on the canvas. Open a family to see its components and what each one does.
Yassine Elhallaoui
Co-Founder, CEO & CTO
Norway
Tap photo to read profileMachine learning, AI, and mechatronics for the real production floor.
Yassine is an AI and machine learning engineer with more than 18 years of experience bridging demanding physical operations and advanced artificial intelligence. He has led field teams of over a hundred people in extreme conditions across global seismic exploration, and has since spent a decade building the automation and machine learning systems that run alongside that kind of work.
Several of those systems went into industrial service. An automated calibration system assesses signal quality in real time and predicts sensor health and degradation, with a companion electro-mechanical rig that applies the same method to geophone cable testing. A machine learning toolkit handles seismic data quality control, cleaning and reporting; a GIS and computer vision interface performs risk analysis and electronic scouting from satellite imagery; and a real-time navigation system guides seismic vibrators in the field while checking the quality of the data as it is acquired. A multi-agent architecture drives system, software and integration testing using computer vision, an end-to-end agentic simulator and digital twin lets a full acquisition workflow be validated without physical hardware, and a corporate agentic architecture consolidates operational data across a company into decision-ready reporting.
His approach centres on agentic AI, large multimodal models, and rigorous uncertainty quantification: explicit confidence thresholds and fallback behaviours, so that autonomous systems are not only capable but safe. He serves as an AI ambassador and contributes to international AI policy work, arguing for security and ethical evaluation to be built into the architecture rather than bolted on later. He has published papers on industrial AI and automation solutions and presented them at industry events internationally, including at ADIPEC, before audiences that judge a method on whether it survives operation rather than review.
He holds a Master of Science in Machine Learning and AI with distinction, and a Doctorate in Business Administration on big data for SaaS business solutions in Industry 4.0 and 5.0. That research kept returning the same finding, and it is the reason this platform exists: the science works, but the infrastructure around it is the bottleneck.
Matthias Assmann
Co-Founder & COO
Norway
Tap photo to read profileTwo decades turning enterprise technology into results a business can measure.
Two decades of leadership across finance and technology. He leads strategy for enterprise data, resource-planning and security programmes, and spent six years before that as an enterprise and integration architect: the person who decides how large systems meet, which is what decides whether a platform is adopted or quietly shelved. He has led teams of business analysts, project managers and developers through those decisions.
His earlier career joined delivery to finance, implementing financial back-office systems, working as an international finance manager, and holding corporate finance, merger and global product-cost roles across European industry. That combination is unusual and it is the reason he owns this side of the company: he has been the buyer, the architect and the person answerable for the number, and he knows which questions each of them asks.
On AlkhemYst-Ai.cloud he owns operations and the commercial side. His aim is a platform that is not only capable for researchers but sustainable, compliant and able to scale responsibly across institutions: research technology an organisation can actually run, govern and account for.
Diana Domanska
Co-Founder & CSO
Norway
Tap photo to read profileA mathematician who builds the methods, then checks what they are allowed to claim.
A mathematician and computer scientist at heart. She bridges the gap between advanced computational engineering and human biology. Holding a PhD in Informatics, a 5-year Master's in Mathematics, and a Bachelor's in Informatics, she brings over a decade of quantitative precision to genomics. Her work as a senior scientist is defined by a fundamental boundary: she engineers methods rather than simply applying them.
Across her career, she has pioneered breakthroughs in colorectal, metastatic liver, breast cancers; ulcerative colitis and other diseases. She integrates complex multi-omic architectures unifying Whole Genome Sequencing, bulk and single-cell RNA-seq, spatial transcriptomics, proteomics, and metabolomics to unlock precision insights into disease progression. Her engineering footprint spans reproducible pipelines for longitudinal transcriptomics, analytical frameworks for spatial multi-omics, genome-wide statistical models, and the interactive visualizations that make complex data interpretable. Building the instrument, rather than simply running it, is what reveals where it breaks.
At AlkhemYst-Ai.cloud, the science stops with her. She curates the platform's method library, validating algorithms against biological literature ground rather than relying on code implementation. Because a plausible wrong answer is the single most dangerous failure mode of modern research tools, her judgment serves as the platform's definitive shield. She ensures AlkhemYst-Ai delivers what real science demands: reviewer-proof analyses, self-evidencing results, and complete auditability that stands the test of time.