PR / Projects / glycocam
In developmentGlycoCam
A carbohydrate counter in development for people living with diabetes, using embedded Gemma 4 AI to support food interpretation.
Project overview
The product focuses on identifying food from images and helping users understand estimated carbohydrate content. It is a software development project, separate from our type 1 diabetes treatment research. Carbohydrate estimates require user review; clinical accuracy and a finished product are not claimed.
How the project works
The intended workflow starts from a meal image, uses embedded Gemma 4 AI to interpret the food and presents estimated carbohydrate information for user review. The aim is to make everyday carbohydrate counting more understandable.
- Image-based food interpretation
- Embedded Gemma 4 model integration
- User-reviewed carbohydrate estimates
PR / 01
The everyday problem
People living with diabetes may need to evaluate the carbohydrate content of meals repeatedly throughout the day. Food composition, serving size and mixed dishes can make this task difficult. GlycoCam aims to support that everyday estimation process with an image-based, understandable interface. The user remains responsible for reviewing the interpretation and the portion information rather than relying on an unexplained single number.
PR / 02
Intended meal-to-estimate flow
The planned experience begins with an image of a meal. Embedded AI interprets likely foods and presents an estimated carbohydrate result with information the user can review. Portion uncertainty and ambiguous foods need to remain visible in the interface. A simple meal and a mixed dish may require different levels of user correction; the design should support that distinction instead of suggesting that every image is equally reliable.
PR / 03
Why embedded Gemma 4
Gemma 4 is the model family specified for the project’s embedded AI direction. The development work needs to balance model capability, response time, memory use and the intended device environment. Embedding a model does not by itself establish clinical accuracy or guarantee a fully offline product. Device support and the final processing architecture will be described when they have been validated.
PR / 04
Validation and user review
Image interpretation and nutritional estimates must be assessed against suitable reference information before strong accuracy claims can be made. Development needs to account for lighting, occlusion, food preparation and portion differences. The product should help users understand uncertainty and correct mistakes. It is being developed as a carbohydrate-counting aid; it is not presented as an autonomous insulin-dose calculator or a clinically approved treatment.
PR / 05
Development and research boundaries
GlycoCam is in development. Public release, supported devices, accuracy evidence and product-specific privacy arrangements will be announced as they are confirmed. It is a separate software project from Punasaari Research’s type 1 diabetes cell-based research proposal. Questions about the product can be sent through the project’s contact link, without submitting medical records or meal images to this corporate website.
Development & availability
GlycoCam is in development. Its estimates are not a substitute for medical advice or an insulin-dosing decision. Clinical accuracy, medical-device approval and a public release are not asserted. It is a separate software product from the diabetes research proposal.