- My role
- Application development and model integration.
- Timeline
- December 2025 - January 2026
Connecting a questionnaire to a model.
GlucoSense is a web-based diabetes risk assessment project. A four-step questionnaire collects lifestyle and biometric inputs and passes them to a prediction model.
The engineering challenge was connecting a short, understandable form to a model that expected a different input structure.
What I worked on.
I built the web application and connected its questionnaire to the prediction API, including the translation between user inputs and model features.
Decisions behind the interface.
Bridging different input formats
The interface asked 17 questions, while the model expected 21 features. A translation layer calculates BMI from height and weight, maps age to categorical buckets, and fills unasked features with default values.
Keeping validation at the boundary
The API validates incoming answers before preparing them for inference. This separates the interface’s form fields from the representation the model expects.
Understanding the trade-offs
Defaulting unanswered features and changing the decision threshold affect the model’s output. The project used a threshold of 0.3; that implementation choice is not evidence of clinical accuracy.
The finished flow.
The application connects a four-step form to the prediction service and displays a risk categorization. The screenshots show the questionnaire and supporting interface.
A closer look.
What this project highlights.
A working model integration needs a clear contract between the interface, validation, and inference. Translating inputs is part of the application design, not just a final connection step.
This is a software project, not a clinically validated diagnostic tool. Its output should not be presented as a diagnosis or a substitute for professional care.