About me
Hi, I am Marcin, a 24-year-old from Katowice, Poland, and an Informatics major from the University of Edinburgh. My to-date academic pursuits focused on internalising as many data analytic tools as possible. This extends to studying disciplines in which the aforementioned tools are mostly applied, including Computer Vision (CV), Natural Language Processing (NLP), Reinforcement Learning (RL) and Time Series Forecasting. I seek positions that concern working with data to extract valuable insights into the real world. Consequently, my professional interests span the Data Science and Machine Learning (ML) fields.
Interests
- Data Science
- Artificial Inteligence
- Machine Learning
- Deep Learning
- Natural Language Processing
- Software Engineering
Education
Master of Informatics with Honours
The University of Edinburgh
2022
I have completed a five-year Integrated Master's program worth 300 ECTS credits at the University of Edinburgh. My alma mater is among the top 15 world universities at the time of my graduation, according to the QS ranking.
International Baccalaureate Diploma
Prywatne Liceum Ogólnokształcące im. Melchiora Wańkowicza
2017
In high school, I challenged myself by rejecting the standard Polish Matura and joining the alternative IB programme. The choice was difficult as I am a Polish native, while IB offers teaching fully in English. In the end, read more... I completed the programme with a maximum possible grade of 7 in 4 of my subjects (Mathematics, Physics, Business and Management, Polish Literature), and grade 6 in the remaining 2 (English and Chemistry).
In my prior education, I have always been a high performer. In my secondary school, I was a laureate of three distinct Regional Subject Olympiads in the Silesian Voivodship, namely in Mathematics, Physics and History. Show less...
Skills
Data Science
Python with its scientific programming toolkit is my favourite programming language. I use scientific Python whether it concerns directly extracting statistical summaries from data, or pre-processing it for downstream read more... machine learning tasks. I have worked with Computer Vision, Natural Language and Time-Series datasets. On occasions, my friends asked me for help when their projects involved data analytic tasks in fields as distinct as Physics and Linguistics. Generally, you name a field with data, and I got your results. Concerning technical quirks, I naturally look for ways to vectorise Python code to get performance speed-ups. In the field of data science, I especially love data visualisation techniques. Show less...
Deep Learning
PyTorch
In my studies, using NumPy, I have implemented from scratch (forward and back propagation) most canonical components of modern neural networks from simple multi-layer perceptrons to complex convolutional layers, recurrent networks, read more... batch normalisation, residual connections, etc. Due to the above first principles work, I have strong background concerning the deep learning concepts and the inner workings of higher-level libraries such as PyTorch.
Reading Andrej Karpathy's The Unreasonable Effectiveness of Recurrent Neural Networks blog post, inspired me to write my own modern PyTorch version of the post's old Torch code base. In the process, I have learned PyTorch and have trained character-level RNN language models that "hallucinated" prose of seminal Polish writers such as Henryk Sienkiewicz.
Later I polished my PyTorch skills during university courses. I have used PyTorch in Computer Vision (CV), Reinforcement Learning (RL) and Natural Language Processing (NLP). In NLP I have also worked with spaCy and the Transformers library of Hugging Face. My most ambitious deep learning project to date is titled: Is RGB all you need? The usefulness of depth in semantic segmentation. Please check its devoted section for more details. Show less...
Web Development
JavaScript is my second language of choice after Python. I have experience in full-stack development. In the front-end, I have worked with vanilla JavaScript enhanced by the DataTables library at Almar IT. On the other hand, at Konsept I read more... have used React. The Python Flask library is my go-to choice for the server side when building web apps. Yet, at Almar IT, I have also used the Java Spring Boot framework in the development of a CRM based on the MVC architecture. Show less...
Natural Language Processing
NLTK
spaCy
Transformers
In the field of Natural Language Processing (NLP), my university has offered me multiple courses such as Processing Formal and Natural Languages, Text Technologies for Data Science, Accelerated Natural Language Processing, read more... Natural Language Understanding, Generation, and Machine Translation (NLU+). As a result, I have extensively studied NLP basics and specialised methods devised to solve the discipline's subproblems. Especially, in the NLU+ course, we have explored the state-of-the-art approaches to NLP, such as attentional models, transformers, BERT or even the brand new pre-trained model prompting paradigm. In practical terms, I have utilised PyTorch with the modern spaCy and Hugging Face's Transformers libraries. In the end, it is worth mentioning that I had the privilege to learn from the world's top researchers in the field. For example, professor Alexandra Birch, one of the inventors of the byte-pair encoding (BPE) method used virtually everywhere in NLP for text segmentation, is also an NLU+ course lecturer. Show less...
Databases
MongoDB
PostgreSQL
MongoDB is my first database of choice. I have designed schemas, built and optimised indexes, and created queries for it. I have used MongoDB during my work at Konsept. Also as part of the Text Technologies for Data Science read more... course at my university, I was involved in the Lyrix project, where we created a search engine for song lyrics. The project involved information retrieval from a collection of over 2 million songs and used MongoDB in the system back-end.
I am familiar with traditional relational databases. Both as part of my university courses and my work at Almar IT, I had to design and instantiate table schemas, and write SQL queries. Show less...
Version Control
Git
GitHub
I have used Git as VCS in my own and university group projects.
UNIX
Bash
Linux
The computing infrastructure at my university was based on Linux machines, so I had to learn UNIX and GNU Bash script. I use Windows on my personal machines.
Virtualisation
Vagrant
VirtualBox
When I must use a library not supported on Windows, I use Vagrant with VirtualBox to set up a VM with appropriate dependencies. In such scenarios involving Python, I like to set up a remote Jupyter Notebook server on the VM, but read more... interact with it within the browser on my host machine. Show less...
Other
The above list includes the remaining programming languages I have used in my university projects.
Haskell was the first language I learned, and to this day I have a read more... good grasp of the functional programming paradigm. Show less...
Blockchain Technologies
Ethereum
Solidity
In my final academic year, I have undertaken a Blockchains and Distributed Ledgers course. In the course, I have studied the core principles behind the technology, and its instantiations in the form of Bitcoin and Ethereum. In read more... practical terms, I have designed, implemented in Solidity, optimised for security, fairness and gas fees, deployed, and interacted with Ethereum smart contracts. My final A1 (95%) grade in the course shall inform one about my "Crypto" concepts comprehension level. Finally, inspired by the course I have implemented a distributed version of the game of chess, using the Ethereum smart contracts. Show less...
Experience
Software Engineer
Konsept App
2021
Konsept by Nagne Studios start-up is a project aimed at delivering a web application that simplifies and automates the recurring parts of the product manager's (PM) job. I was directly involved in designing app features read more... and detailing their implementation possibilities, given my machine learning background. A major feature of the Konsept App in the design of which I was involved concerned a recommender system. To create such a system, we had to build a database of items for recommendation. I have gathered such a dataset by implementing a custom web crawler and data parser. The next step concerned data annotation, and for this purpose, I have built a proprietary web application. Concerning my programming involvement, I have used Python, Flask, React, JavaScript and MongoDB during my work at Konsept. Show less...
Shop Assistant
Tartan Weaving Mill
2019 - 2020
Tartan Weaving Mill is the largest souvenir shop in Edinburgh. In the shop, I was directly involved in customer service.
Intern Software Engineer
Almar IT
2018
Almar IT is a medium-sized company delivering CRM systems to petroleum distributing businesses. During my internship, I was introduced to the company's large code base. The involved technological stack primarily included Java, Spring read more... Boot, Hibernate, PostgreSQL/Oracle database, JavaScript, and DataTables library. In order to contribute, I had to learn Spring Boot and JavaScript from scratch. Finally, I was involved in developing a feature of the main CRM product. Show less...
Languages
Polish
I am a native speaker of Polish.
English
Since high school, all of my education was undertaken in English. I have lived, studied and worked in the UK for 5 years. My grade 6 in English B HL of the International Baccalaureate Diploma is a certificate of my English language proficiency at the B2+ level (reference).
German
I hold a Goethe-Zertifikat A1.
Projects
Is RGB all you need? The usefulness of depth in semantic segmentation.
As part of the Machine Learning Practical course, together with Maciej Kowalski, we have approached a Computer Vision (CV) problem of utilising non-RGB data to improve the performance of existing classifiers. Specifically, we have read more... attempted to outperform the RGB-only baseline model on the dataset of drone-captured aerial landscapes, by incorporating the also available per-pixel depth/elevation information recorded with the LIDAR sensors. The task concerned semantic segmentation, so assigning a label to each pixel of an image, to determine whether such a pixel is a part of a house, road, car or tree, etc.
We have investigated different ways of augmenting the existing ResNet backbone architecture to add depth information. We have also aimed to utilise the off-the-shelf pre-trained models rather than training new ones from scratch. A meaningful introduction of untrained data channels into already pre-trained architecture posed a grand challenge. In the project, we had to schedule our hyper-parameter tuning experiments on the available university GPU cluster. Additionally, one of the solutions required me to overwrite the existing convolution operation to make it "depth-aware". In the process I have implemented a custom PyTorch convolution module, using basic vectorised mathematical operations. The module is available on my GitHub.
The final project results can be found in the attached pdf report. Show less...
WebSweeper
To learn JavaScript and Bootstrap I have written from scratch a responsive Minesweeper game web application. Later on, I enhanced the project by adding a solver that computes the probability of a bomb being present at a tile read more... in the game and hence, giving a hint to a player. I have designed the solver entirely myself, by using my knowledge of probability and analysing the process in which a human plays the game. Naïve solutions often lead to exponential blow-up in terms of application memory requirements, so I had to look for clever ways to overcome the problem. Show less...
EthereumChess
Inspired by the Blockchain and Distributed Ledgers course I have implemented a distributed version of the game of chess, using the Ethereum smart contracts.