Michał Gacka
ML Software Engineer

I'm a machine learning engineer and a full-stack web developer. I believe that technology should serve humans and not the other way around. I want to use machine learning to improve quality of life and solve important problems of today.

Selected Projects

Protein tertiary structure prediction

Predicting the tertiary (3-dimensional) structure of a folded protein given its primary structure (sequence of amino acids).

Product recommendation and churn in insurance

A product recommendation system and churn model for insurance companies.

Visual attention applied to object recognition

CNN's attention compared to human attention based of eye-tracking data.

Endoscopy Quality Evaluation

Software that created a real time map for assesing quality of a colonoscopy exam. The project was developed as a desktop app based on gastroentherologist's expertise.

Parking Occupancy Detection

Real-time detection of parking space occupancy. The project was a POC that aimed to replace expensive sensors. I used traditional computer vision techniques.

TRAiVEL - Sentiment-based Travel Recommendations

An app that uses sentiment analysis to estimate current mood in European countries based on twitter data and recommended travel destinations together with flight prices fetched from Skyscanner api.

Parametric t-SNE for temporal Emotion Recognition visualization

A novel technique for visualizing emotion recognition results including the temporal dimension. An "emotion space" is created using parametric t-SNE, where machine learning models can be visually compared with respect to modalities, models, and hyperparameters used.

Rule-based Insurance Recommendation - Admin Panel

A web app that allows its user to set rules that get evaluated based on input features. Its primary goal was to create a universal way of deciding which insurance products should be displayed depending on the user's information.

CNNs & Transfer Learning for Furniture Recognition

Project developed for a course and a Kaggle competition. I used Convolutional Neural Networks to distinguish between 128 classes of furniture.