Machine learning with EEG brainwaves and pupil diameter to infer real-time productivity, supported by a backend for live data and a frontend for visualizing and competing with friends.
MindsEye is a neuroscience-driven application that combines EEG brainwave data and eye-tracking measurements to estimate focus levels in real time.
The system enables users to participate in competitive productivity sessions with friends, visualized through a responsive web dashboard.
| Category | Description |
|---|---|
| Machine Learning | XGBoost model trained on EEG and pupil data to infer productivity levels. |
| EEG and Pupil Tracking | Integrates Muse EEG headset and MediaPipe iris tracking for synchronized brain and eye data collection. |
| Backend | FastAPI-based server handling ML inference, data storage, and WebSocket communications for live competitions. |
| Frontend | React-based interface that visualizes focus scores, leaderboards, and personal statistics. |
| Calibration | Each user runs a personalized calibration session to establish baseline focus levels. |
| Real-time Competition | Users can connect with friends in live sessions and compare focus scores as they work. |
frontend/ → React app for EEG streaming, visualization, and interaction backend/ → FastAPI service for data handling, prediction, and live communication models/ → Machine learning models (XGBoost) and preprocessing scripts
System Flow
- The frontend captures EEG and pupil data.
- Data is streamed to the backend via API requests.
- The backend uses a calibrated XGBoost model to infer productivity.
- Predictions are broadcast to all connected clients.
- The frontend visualizes productivity levels and competition results in real time.
Built with React, the frontend collects EEG and pupil data, visualizes real-time focus levels, and manages live competition sessions through WebSockets.
It includes calibration, competiton data, and productivity trend views.
Developed using FastAPI, PocketBase, and WebSockets, the backend handles data ingestion, model inference, and live synchronization between users.
It exposes endpoints for calibration, prediction, and session management powered by the trained XGBoost model.
- Algorithm: XGBoost regression/classification
- Inputs: EEG channels (AF7, AF8, TP9, TP10) and pupil diameters (left/right)
- Output: Productivity score between 0 and 100
- Calibration: Each user establishes an individual baseline before competition
- Adaptation: Model recalibrates periodically to adjust to signal drift
| Layer | Tools |
|---|---|
| Frontend | React, WebSockets |
| Backend | FastAPI, Python, XGBoost, asyncio |
| Database | PocketBase / SQLite |
| ML / Data | XGBoost, NumPy, Pandas |
| Data Input | Muse EEG, MediaPipe Iris Tracking |
| Hosting | Render (backend), Vercel (frontend) |
