top of page
GLUCOSE ML
A data platform solving the biggest bottleneck in diabetes AI research by streamlining discovery & evaluation of CGM datasets.
Team: Lead Research Professor, 2 Software Engineers, 1 Data Engineer, 1 Qualitative Researcher, 1 Designer (me).
Timeline: 4 months (ongoing)
Tools: Figma, Miro, Claude Design, Typeform
Outcomes:
-
Launched 0->1: 3 full iterations from info architecture to metadata schema to final UI.
-
Validated with full HCI user study: robust testing of core flows that drove each redesign.
-
Submitted to CHI research publication 2027: findings on ML dataset repository needs and biases.
Role: As the Product Designer (nights/weekends alongside full-time role), I led 0 to 1 end-to-end design, testing & launch of the web platform. From user discovery to info architecture to metadata schema to UI to full HCI user study for ACM CHI publication.
RESULTS
3-4 weeks -> 10 minutes to access and evaluate a dataset
1 -> 5 datasets evaluated on average before choosing
0 -> 50+ users in 1 month
CONTEXT
Continuous glucose monitoring (CGM) is one of the richest signals in metabolic health that powers AI health research and products. These minimally invasive sensors now stream real-time glucose data and are becoming an everyday wearable for millions.
USER
PROBLEM
Finding and vetting a CGM dataset is so tedious that researchers settle for the first one they find.
1. Discovery is manual & scattered
Researchers burn weeks searching across academic papers, clinical portals, and one-off repos.
2. Evaluation data is
inconsistent
Different metadata, access rules & documentation make it impossible to compare without downloading each one.
3. ML quality gets compromised
Settling for the first dataset means biases slip through & model quality suffers downstream.
BUSINESS
IMPACT
As adoption spreads to everyday consumers, the volume and value of CGM data compounds.
Improving how researchers find & vet data raises the quality on every product built downstream. CGM is a ~$16B market growing 15% a year and one of the fastest growing health AI frontiers.
How might we help researchers discover, compare, and access the right CGM datasets in minutes?
Reduce time-to-dataset discovery
Increase # of datasets compared & reduce bias
Increase choice confidence & ML output quality
SOLUTION
Glucose ML
A centralized CGM dataset hub built around a three-step evaluation flow:
2. Compare
Determine compatibility by viewing dataset characteristics side by side.
1. Discover
Browse & filter for dataset granular variables based on research goals.
3. Evaluate
Assess final fit through in-depth visualizations, metadata, and one-click access to download.
⚡ Case study in progress - updated daily!
More projects:
Let's connect!
If you like what you see, I'd love to chat!
Alternatively, here's a fancy sheet of paper:
bottom of page