Etele Kovács

Research tool · Python · Cellpose · OpenCV · SLURM

• Turned DAPI fluorescence microscopy into cell counts and biovolume (µm³): fine-tuned Cellpose on the ITU HPC cluster (SLURM + Apptainer) and ported the reference Zeder biovolume algorithm from the field-standard tool's decompiled C#.
• Shipped as an offline desktop app with live microscope capture and CSV export — in daily use by researchers.

Python · PyTorch · OpenCV · Computer Vision

• Built a 6-channel Faster R-CNN that detects and counts pico-algae in paired microscopy images; self-annotated 16,181 objects in CVAT and tuned post-processing to cut count MAE from 3.94 to 2.42.
• Shipped as an offline desktop app — installer, interactive review editor, live microscope capture — whose corrections export straight back into training data. In daily use by researchers.

Group project (ITU) · Python · PyTorch · Transformers · SLURM

• Four-person NLP course project with a paper: built the training/evaluation pipeline and a language-aware bias layer that lifted XLM-RoBERTa from span F1 0.636 to 0.649 on the LINCE benchmark (67K tweets), with SLURM cluster training.

Group project (ITU) · Python · NumPy · scikit-learn · PyTorch

• Three-person BSc exam project on ~678K French motor policies: built a neural network from scratch in NumPy, its PyTorch reference, PCA, and the preprocessing/EDA pipeline; a Negative Binomial GLM won (test RMSE 3.03) — on noisy, zero-inflated data the right statistical model beat the complex ones.