01 — Work

Experience

I've worked in research and in industry. Here's where I've been.

Timeline

newest first

Oct 2025 — present

Research Assistant

University Hospital Carl Gustav Carus · Cognitive Neurophysiology, TU Dresden

Architecting a spatiotemporal graph neural network (ST-GNN) to classify ADHD from multi-channel EEG.
Engineered a custom phase-locking-value prior to initialize graph connectivity — so the model learns from functional synchronization rather than physical electrode distance.
Built a hierarchical transfer-learning pipeline: pretrain on large-scale trial data, fine-tune on patient-level diagnosis to overcome data scarcity.
Developed a differentiable patient aggregation layer fusing mean neural signatures with intra-individual variability metrics.
Ran a formal MANCOVA variance decomposition to isolate the graph's inductive bias; achieved >70% predictive accuracy on unseen subjects.

GNNEEGPLVtransfer learning

Nov 2025 — Jun 2026

Master's Thesis Student

DZNE · BIONICS lab, Dresden

Developed a high-performance computational pipeline for large-scale neural dynamics from high-density microelectrode arrays (HD-MEA, 4,096 channels).
Engineered a dual-stream signal-processing framework — extracting LFP for synaptic-input analysis and spiking output.
Implemented advanced spectral parameterization, plus features quantifying spatial entropy and burst-propagation trajectories.
Designed an unsupervised, label-free clustering protocol (PCA, UMAP, GMM) with strict leave-one-slice-out validation against small-sample overfitting.
Optimized data ingestion for high-velocity recordings via vectorized artifact rejection and Gaussian kernel smoothing.

HD-MEAMNE-PythonclusteringHPC

Dec 2023 — Jul 2025

Co-founder & AI Lead

Fintech startup · Dresden / remote

Where I built product-level ML: fraud detection, credit-risk and churn systems; feature engineering and automated model selection; and explainable AI (SHAP, LIME) to make model decisions defensible. Reached an 80% reduction in false positives over the incumbent rule-based system.

fraudcredit riskXAIproduct ML

Dec 2021 — Sep 2023

AI/ML Engineer & Data Scientist

Wipro Technologies · Financial Services, Bangalore

Where I learned to ship machine learning for real: end-to-end fraud-detection pipelines using neural networks and ensembles; NLP for document analysis and automated scoring; scalable data-processing on Azure ML and Synapse; and statistical hypothesis testing for model validation.

end-to-end MLNLPAzurevalidation

Across roles I've worked on the whole path — from lab recordings to production systems — and I try to apply the same care everywhere.