Data Mesh Engineer | Licenced UAV Pilot | Artful Combat Explorer | Programmer by profession | Versatile Fit Adventurer | Whiskey Dandy | Geek Realm Gamer | Nature Hunter
My background starts at the Faculty of Informatics and Information Technology at Slovak Technical University. where I successfully completed my bachelor's degree. My thesis focused on the application of deep learning in the medical domain. During my time at this institution, I touch almost every aspect of computer science and gained proficiency in diverse programming paradigms. The curriculum include a comprehensive range, from understanding the logic of single and multiple register CPU, low-level programming languages such as C, networking, Unix kernel utilization, bash scripting, and database systems, to fundamental principles in artificial Intelligence, web technologies, higher-level programming languages, and User Experience (UX) design.
After that, I directed my focus towards Artificial intelligence, where I studied the field of Artificial intelligence and processing of Big Data at Masaryk University in Brno. Here my master's thesis was oriented around the broader applications of Artificial intelligence in the real world within IT companies. This academic phase provided an extensive exploration of AI covering areas such as state space search, genetic algorithms, outlier detection, Natural Language Processing (NLP), deep learning, information retrieval, machine learning, among others. The curriculum also extended beyond AI, incorporating database systems, encompassing both relational databases (with a focus on optimization and efficiency) and NoSQL databases like Redis, Mongo, Elastic, etc... Here I gained insights what are the main duties of data-oriented roles understanding data warehouses, cloud computing and the architecture of supercomputers.
Towards my academic pursuits, I searched for practical experience in the professional world. This led me to have part-time jobs where I took on roles such as a software engineer, AI developer, ETL pipeline specialist, and Data engineer. These experiences have been instrumental in reinforcing and applying the theoretical knowledge acquired during my academic journey.
AutoPlateMonitor combines the user-friendly PyQT GUI with the robust capabilities of OpenCV for real-time license plate recognition. Experience a live camera feed as the application dynamically identifies and segments license plates in the background. Users can manually input plate details and receive instant notifications upon detection. Perfect for personalized arrival alerts and security setups. Streamline your license plate recognition with AutoPlateMonitor.
Exernote, powered by Ruby on Rails, is a dynamic web application designed to revolutionize your fitness journey. Effortlessly track and manage your gym progress, keeping detailed records of exercises and sessions. Gain insights from basic statistics showcasing your achievements. Key features include global accessibility via any web browser and any device, support for multiple languages, robust authentication for data security, and efficient data storage with PostgreSQL. Exernote empowers users with a seamless and personalized fitness experience. Elevate your workout routine with Exernote today!"
Dive into our school project featuring a Feedforward Neural Network implemented in C++, leveraging the backpropagation algorithm and optimized with gradient descent with momentum. Developed without the aid of third-party libraries, this collaborative effort aimed to create a neural network capable of classifying Fashion MNIST images with over 90% accuracy. The project successfully trains and evaluates within 30 minutes, using a dataset comprising 60,000 training examples and 10,000 test examples, each a 28x28 grayscale image associated with one of 10 classes. Delve into the code and discover the intricacies of our neural network solution for image classification in the CSV format dataset.
Python application that leverages the LangChain library to harness the capabilities of the LLAMA-CPP model, enabling you to build your personalized chat bot. Beyond conventional chat features, this application delves into diverse linguistic functionalities, simplifying coding and scraping tasks. Notably, LLAMA-CPP's Quantization enables smooth execution on both CPU and metal (Apple M1)
CloudInventario is a robust work project designed to monitor various cloud providers, offering users insights into their resource usage across different platforms. The heart of this project is a visually stunning portal application, built with the Python Dash library and featuring beautiful visualizations using Bootstrap components. Users can effortlessly navigate through managed resources, view detailed invoices, and stay informed about their cloud usage. Key functions include monitoring a wide range of cloud resources, such as virtual machines, load balancers, SQL servers. It is integrated with major cloud providers, including GCP, AWS, AWS Lightsail, Azure, Hetzner Hcloud, Proxmox, and many others. The Python Dash library enhances user experience by creating visually appealing and dynamic visualizations, providing real-time insights into cloud usage. Periodic database refresh, facilitated by Jenkins and load balancing, ensures users are presented with the latest and most accurate information.
Sophisticated job portal scraper meticulously designed to curate a comprehensive database of current job listings.Featuring custom middleware to efficiently handle multiple proxies, the scraping process is orchestrated through a configuration file. The ETL pipeline, a core organizational structure of the project, encompasses various stages. Starting with scraping entire pages for job listings, the subsequent steps involve modifying content. This includes language identification, translation into specific languages, summarization of content (both HTML and plain text), extraction of specific fields, and finally, loading the data into a relational database. Each scraping portal is equipped with its own spider, ensuring optimized results for search fields. This project seamlessly combines the power of Python, Scrapy, and ETL methodologies to deliver a comprehensive solution for creating and managing a dynamic job database
Embark on a revolutionary experience with our Python work project, harnessing the robust capabilities of OpenAI to craft an advanced chat bot. This innovative application extends its functionality to include real estate transactions, allowing users to seamlessly explore and purchase houses and other properties. Leveraging OpenAI, the project enables users to perform natural language searches on reality pages, parsing the results into JSON and effortlessly finding their desired homes
Classification of semiconductor chip layers in a FIB-SEM microscope. In my master thesis, I collaborated with an external company specializing in the soldering of semiconductor chips with defects, aiming to identify the specific layer where errors occurred for further inspection. Given the minute scale of the chips, only a few nanometers, a focused ion beam was employed to desolder the layers, and their content was examined using a scanning electron microscope. With the objective of improving accuracy and expediting the process, I endeavored to invent an algorithm capable of detecting the layer within the chip efficiently. Despite a limited dataset, I explored a mix of general algorithms, favoring practicality over deep learning. My implemented alternatives encompassed clustering, neural networks with smaller architectures, template comparison, and adaptive thresholding. The method I devised, utilizing white/black hat for data cleaning, structure tensor for edge detection, and computing histograms of inflexion points, outperformed others, showcasing superior results.
Processing of volumetric medical data to support medical diagnostics via deep learning a.k.a. Processing of image data using artificial intelligence methods. In my bachelor thesis, I delved into the realm of processing volumetric medical data to support medical diagnostics through the application of deep learning, specifically the processing of image data using artificial intelligence methods. My focus centered on comparing different approaches—manual and automatic—of creating kernels in convolutional neural networks. Employing the Unet architecture, the task involved classifying images to determine the presence of tumors. The dataset encompassed images of breast cancer and histological images of cancer in the kidney. For comparison, I explored classical approaches like creating kernels through backpropagation, leveraging autoencoders to simplify images for subsequent classification, and implementing transfer learning by utilizing weights from different models. The final approach involved a manual touch—using a Gaussian function to create a filter for the kernel. Remarkably, the best results emerged from a hybrid approach, manually defining a kernel from a Gaussian function that was later refined, albeit not entirely overridden, through backpropagation iterations
