03 / 05
Experiments
Projects / shipped, local, and on the bench.
Two deployed, one built, one planned. Status is labelled honestly.
SIM
Motor-imagery topography.
This visualization replays a motor-imagery trial: baseline, cue, then imagined movement. During right-hand imagery, C3 over the left motor cortex shows ERD (desynchronization). During left-hand imagery, C4 over the right hemisphere does. This is the canonical signal the projects below try to read robustly, across people.
- Montage
- 10–20 · 19 channels
- Sampling
- 256 Hz simulated
- Band
- μ · 8–12 Hz
- Paradigm
- Cued motor imagery · 2-class
- May — Jun 2026DEPLOYED
eeg-mi-benchmark
Live benchmarking platform · CSP+LDA, Riemannian
- Full-stack benchmarking platform for motor imagery classification. FastAPI backend preprocesses BNCI2014001 (9 subjects, 22 channels, 250 Hz) via MOABB at build time, serving time-series, PSD, and MNE-rendered scalp topoplots.
- Runs CSP+LDA and Riemannian+LDA classifiers in parallel across all 9 subjects, streaming fold-by-fold results to the browser via Server-Sent Events. Results include accuracy with confidence intervals.
- Next.js frontend with URL-persistent state (dataset, subject, run, epoch), epoch navigator, and annotate mode. Deployed on Vercel + Railway.
Next.jsTypeScriptFastAPIMNE-PythonMOABBPyRiemannscikit-learneeg-mi-benchmark.vercel.app - May — Aug 2025BUILT
CalmPulse
Real-time ECG biofeedback system
- Designed and implemented a low-cost ECG-based stress monitoring device using Arduino Uno (3-lead setup, amplifier/filter circuitry).
- Built full-stack system with React (shadcn) frontend, Socket.IO backend, and live ECG streaming with QRS peak detection for BPM calculation.
- Developed adaptive breathing guide and zone-based heart rate classification (Relaxed/Elevated/Anxious) with session logging and visualization.
- Delivered functional prototype integrating hardware, backend signal processing, and interactive UI for stress management training.
ReactTypeScriptSocket.ioExpressArduinogithub.com/muhaiminsarker/pulse-zen-tracker - Spring 2025DEPLOYED
EEG Motor Movement/Imagery Analysis
JHU Biomedical Data Science capstone · PhysioNet
- Capstone project for JHU EN.585.771 Biomedical Data Science — interactive Streamlit app for exploring the PhysioNet EEG Motor Imagery Dataset (109 subjects, 14 runs, 64 channels, 160 Hz).
- Data pipeline loads EDF files, renames channels to standard 10-20 nomenclature, applies 8–30 Hz bandpass filter, and extracts epochs from −0.5s to 2.0s.
- Three visualization modes: motor cortex time-series (C3, C4, Cz), power spectral density (Mu 8–12 Hz / Beta 13–30 Hz rhythms), and side-by-side run comparison.
PythonMNE-PythonStreamlitNumPySciPygithub.com/muhaiminsarker/eeg-motor-imagery - 2026PLANNED
mindsculpt-web
Mental imagery → live geometry
- Browser-based visual mental imagery tool. EEGNet classifier outputs posterior probabilities across geometry classes; those probabilities drive a Three.js MarchingCubes render.
- When imagination drifts between two shapes, a hybrid form appears. The ambiguity in brain state becomes the creative output.
Next.jsThree.jsFastAPIMNE-PythonBraindecode