The question in AI has shifted from “can we build a model?” to “can we trust it, govern it and put it to work?” DementAI answers all three.
The DementAI team from Katalyze Data, a UK-based consultancy and long-time SAS Partner, was named the 2025 SAS Hackathon Grand Champion at SAS Innovate. Every year, the SAS Hackathon turns a month of collaboration into solutions for real problems.
Their system is designed to help flag Alzheimer’s disease up to two years sooner than existing approaches not by replacing clinicians, but by surfacing the subtle patterns of decline already sitting in patients’ clinical records – and long before a referral to a specialist.
We didn’t build DementAI just to make predictions. We built it to buy patients’ time. Tamás Bosznay, Principal Consultant, Katalyze Data
That’s the goal. What made it achievable in a month? The platform underneath it.
One platform, from raw data to real-world decision
What makes DementAI remarkable isn’t a single, clever model. Rather, it’s that the team took on the entire problem, end to end, without ever leaving SAS Viya.
All in one platform, the team:
- Built models on EEG data using RNNs and LSTMs in Python.
- Modeled tabular health records and applied natural language processing to unstructured clinical notes in Model Studio.
- Blended structured records, brain scans and physician notes, using synthetic data where appropriate to protect privacy.
- Registered everything in Model Manager as a central repository for comparison and monitoring and deployed it for inference – all in one governed environment.
No more stitching together half a dozen disconnected tools, and no data leaving the platform. By building on SAS Viya, the team was able to fully unify data, analytics and governance in a single place.
Governance isn’t the price of speed. It’s the source of it.
In a regulated industry, an AI model is only as valuable as it is trustworthy. The old assumption was that trust slows you down, but DementAI proves the opposite.
What stood out wasn’t only the accuracy of the models, but how the team built them: using governed workflows and privacy-preserving techniques. In regulated environments, that foundation is exactly what makes the leap from prototype to real-world pilot possible.
Explainability, audit trails, bias reporting, ongoing monitoring, full data lineage: on SAS Viya, these aren’t features you bolt on at the end. They’re built in.
Bring your own language
The team moved fast for another reason: nobody had to abandon the tools they knew. SAS Viya lets users work in Python, R and SAS within the same governed environment.
A data scientist who lives in Python and a statistician who has spent a career in SAS can build toward the same trusted, deployable result.
Agentic AI is already in the workflow.
The team wired their models into an agentic AI workflow in SAS Intelligent Decisioning, combining their own machine learning with large language models to transcribe audio, summarize patient history and generate reports. And it’s all kept on the rails by deterministic rules and a human in the loop.
That’s not a future vision. It’s a working prototype, built in a month, on a platform designed to carry an idea from messy raw data all the way to governed decision support.
The stakes are too high for anything less
By 2050, the number of people living with Alzheimer’s is expected to reach 139 million. Behind each of those figures is a person, and the people who love them, watching for the small signs that something has changed.
DementAI is a remarkable achievement. But the real story is the opportunity it represents for any team, in any regulated industry, sitting on data that could help them see earlier. The platform to do it already exists.
The only variable is how quickly you choose to move.
