About
This is a place where I write about what I read and what I work on related to Machine Learning
Last updated
This is a place where I write about what I read and what I work on related to Machine Learning
Last updated
I am currently working in Machine Learning. I have a background in software engineering, research, entrepreneurship.
Formally trained as a software engineer, a programme where I spent 5 years coding various things (operating systems, websites, games, you name it - typical Computer Science track at a Polytechnics university). Spent a few more years afterwards as software developer.
Had the chance to pursue a Research MSc in Cognitive Science. After which I decided a PhD and academia is not the suitable environment for me. I co-founded a hardware company. It wasn't a unicorn, nor did we want it to be. We wanted a regular bootstrapped organic growth company.
Something was unfolding during these years and it looked impressive. AI was becoming bigger and bigger and I wanted to be a part of that. I returned to studies, to upskill: Maths, Statistics, Machine Learning, productionizing ML and the rest. Through a MSc program, a lot of self-study, personal projects and working in a company that does AI for insurance, I became a junior ML Engineer, but with a rich experience in other fields, which I believe constitutes an advantage.
NLP, Computer Vision, tabular data - I enjoy working on either.
Moreover, I choose work conditioned on the product having a benefit to society and the team to be good, passionate, dedicated and fun too.
In my spare time, I do 🚴, 🧗, ✈️, 🍝
R&D work for building a search engine for the insurance domain (questions / answers system)
My work included reading recent literature (most important research papers and a few books along the way) on state of the art in Information Retrieval and developments in the past few years.
I used Transformer based models to build a search engine that would return answers to questions by looking through a corpus of insurance related text.
Choose, fine-tune and customize architecture of Deep Learning models for automatic processing of handwritten information in accident reports.
Created a software library for generating synthetic images that immitate human handwriting and automatically annotate each word with bounding boxes.
This allowed us to grow our training set from 3000 images to 300.000 images and train a model with near maximum accuracy on our word splitting task, which was integrated into the production pipeline for automated documents processing.
