Software & Projects

Statistical ML methods · Research software · EEG & BCI

My core research software examines how models behave when training and evaluation data differ. I built tools to compare transfer strategies, separate model degradation from adaptation and retraining gains, and inspect the decisions behind each experiment. EEG recordings provide the main empirical setting. The full collection includes feature and adaptation libraries, offline diagnostics, simulation, benchmark inspection, and sample-paced pseudo-online replay. Additional projects extend this work into electrode-layout optimization and local vision-and-device integration.

01

Complementary tools for reproducible transfer experiments, offline drift analysis, and sample-paced pseudo-online replay.

CrossDA project cover

Cross-session DA experiment runner

CrossDA

CLI and Python API for comparing transfer-learning strategies with matched experiments, detailed results, and auditable session roles. Evaluated using repeated-session EEG.

Evidence: CLI + Python API; MAP, DWP, MMP, and BDP; auditable summary, detail, and session-role outputs

Details GitHub
ShiftDx project cover

Drift diagnostics dashboard

ShiftDx

Offline analysis of model degradation, adaptation benefit, and retraining gaps across EEG sessions, with fixed-reference comparisons and statistical reports.

Evidence: Five drift metrics; fixed-reference analysis; no-adaptation, adaptation, and retraining comparisons

Details GitHub Live Demo

Pseudo-online BCI desktop app

NeuroStream

Desktop app for sample-paced pseudo-online replay of recorded EEG, with motor-imagery pipelines, Euclidean Alignment, and progressive inference visualization.

Evidence: CSP, FBCSP, and tangent-space pipelines; 0.5 s checkpoints; sample-paced pseudo-online replay

Details GitHub

02

Build stable representations

Train/test-separated whitening and feature extraction for comparable EEG trials.

eegwhiten project cover

EEG whitening and alignment package

eegwhiten

R package for whitening EEG channels or feature matrices with reusable models, diagnostics, cross-session recentering, and Riemannian tangent-space helpers.

Details GitHub
BCIFeatR project cover

EEG feature extraction package

BCIFeatR

R toolkit for train/test-safe EEG feature extraction across spatial-filter, covariance, spectral, dynamical, and avalanche feature families.

Details GitHub

03

Adapt across sessions

Reusable R and Python methods for alignment, transport, and source-target shift analysis.

DA4BCI project cover

Domain adaptation package

DA4BCI

R toolkit for aligning EEG feature distributions across sessions or subjects and evaluating source-target shift before and after adaptation.

Details GitHub
DA4BCI-Python project cover

Python domain adaptation package

DA4BCI-Python

Python backend for EEG/BCI domain adaptation, shift metrics, SPD geometry, preprocessing, and drift detection.

Details GitHub

04

Audit benchmark results

Interactive inspection of source-session roles, configuration sensitivity, and runtime tradeoffs.

MSDA-Bench project cover

Multi-source DA benchmark dashboard

MSDA-Bench

Interactive benchmark dashboard for comparing source-session utilization strategies across MAP, DWP, MMP, and BDP pipelines.

Details GitHub Live Demo

05

Simulation & optimization

Electrode-layout search and model-based comparison, alongside tools for covariance geometry, augmentation audit, and method intuition.

EEGLayoutLab project cover

Simulation & optimization

EEGLayoutLab

An interactive application for electrode-layout design, bounded optimization, and forward simulation. A 3D interface connects to Python computations and reusable local results; evaluations use one simulated head model.

Details
TensorEEG project cover

EEG simulation and augmentation audit package

TensorEEG

R package for synthetic EEG tensor simulation, SPD covariance augmentation, fidelity auditing, and manifest-based replay.

Details GitHub
TensorEEG-py project cover

Python simulation and augmentation audit package

TensorEEG-py

Python package mirroring TensorEEG's simulation, augmentation, SPD geometry, fidelity audit, and manifest replay APIs.

Details GitHub
ShiftLens project cover

Interactive DA method explorer

ShiftLens

Static browser app for teaching and inspecting domain-adaptation geometry with 2-D toy data, method animations, metrics, and export tools.

Details Live Demo

06

Engineering Projects

Experimental application development that connects existing vision models to camera and device-control workflows.

Vision & device-control prototype

GimbalPaw

An experimental macOS application connecting Apple Vision person tracking, camera capture, and BLE gimbal control. The implementation includes motion smoothing, target reacquisition, and stale-frame checks.

Details GitHub

+

Selected EEG Competition Work

Kaggle HMS - Harmful Brain Activity Classification, Silver Medal, rank 98 of 2,767 teams.