Non-Prick Blood Glucose Measurement (AI + IoT) – Student Project Kit

Track glucose trends without a needle. This project shows how optics, biosignals, and machine learning can estimate glucose trends non-invasively—perfect for final-year demos, research fairs, and placements. Build the full stack: sensing hardware, on-device processing, cloud analytics, and a companion mobile app with live dashboards.

Why you’ll love this project

• No needle demo: optical sensing + biosignals to estimate glucose trends
• End-to-end build: embedded hardware, Android app, cloud, and ML pipeline
• Recruiter-friendly: signals, features, models, validation—everything to talk about in interviews
• Polished UI: real-time charts, alerts, and historical insights on your phone
• Ready to extend: add diet/activity notes, sleep tags, or stress markers for richer models

What it does (in simple words)

Place your finger on the sensor. The device shines safe light and records reflections along with pulse and skin parameters. Features are computed on-device and sent to the app. A trained ML model estimates the glucose trend (rising, steady, falling) and a continuous numeric estimate for education and research purposes. You can compare with optional finger-prick readings to improve calibration over time.

How it works (technical flow)

• Sensing: PPG module with multi-wavelength LEDs (e.g., red/IR/NIR), ambient-light rejection, and temperature/GSR support
• Feature Engineering: pulse waveform features (AC/DC ratio, time-domain peaks, slope, area under curve), perfusion index, temperature-adjusted factors, motion quality index
• ML Models: start with XGBoost/Random Forest; optional small Neural Network for regression
• Training: Google Colab notebook with dataset ingestion, cleaning, K-fold validation, and export to TensorFlow Lite
• App: Android (Flutter/Jetpack Compose) shows live stream, trend arrows, sparkline charts, daily/weekly reports, and calibration screen
• Cloud: Firebase/Firestore for logs, model versions, and user sessions; optional offline mode
• Edge: ESP32 (Wi-Fi/BLE) handles sensors, pre-processing, BLE streaming, basic filtering