Guided data flow
The interface turns a multi-stage ML pipeline into a sequence users can understand and recover from.
02 / AI health insights
VitalSight
A report-to-insight workflow that turns uploaded health data into accessible model-assisted predictions.
VitalSight explores how web products can make machine-learning outputs understandable without presenting them as medical certainty.
The challenge
Raw report data, model inputs, and user-facing explanations live in different technical worlds. The system needed to connect them while keeping uncertainty and the limits of an early-awareness tool visible.
The approach
The experience is organized as a guided pipeline: upload a report, extract relevant values, validate inputs, request a prediction, and explain the result with clear boundaries.
02 / System story
One connected product flow.
01Upload report
02Extract values
03Validate data
04Run model
05Explain result
Core stack
ReactNode.jsExpressMongoDBscikit-learn
03 / Key decisions
Designed around the hard parts.
The most important work is often deciding what the system should make obvious, durable, or intentionally limited.
Model behind an API boundary
Prediction logic stays independent from the product interface, making the system easier to extend with additional modules.
Responsible language
Results are framed as early-awareness signals, not diagnosis, and the product directs users toward professional review.
04 / Current outcome
The open-source prototype supports report upload and model-assisted predictions for anemia and diabetes, with a modular path for future conditions.
This case study reports only publicly supportable outcomes. No usage or business metrics have been invented.