UCL student | London Based | AI Research & Development
At the beginning of 2021 I co-founded the Verifee project and have been working on it with O2 ever since. It is an AI engine that provides an explainable analysis of the credibility of news articles through a browser extension that warns users about dangerous articles. Our mission is to make it available to as many people as possible and to help prevent more people to get manipulated by fake news.
Two years long collaboration with Charles university and Václav Moravec on research regarding the overlap between artificial intelligence and journalism, where we worked on tackling the issue of insufficient amount of high-quality data for disinformation analysis and recognition. We established a methodology for article classification that centres on capturing objectivity and elementality, which allows it to be ideal for producing training data for NLP models. Following-up on this, we have produced a large dataset of annotated articles and we are in the process of publishing our paper on this topic.
Collaboration with an Intel and Microsoft Xbox team on creating technology, which aims to allow users to play these games in a more immersive way without any handheld device and in a way more accessible to those with disability. Our mission is to make gaming more accessible and push how games can be played. The whole project is happening via UCL MotionInput iniciative.
As the head of tutorials at UCL Artificial Intelligence society, I have the responsibility of delivering weekly tutorials, which run on a weekly basis throughout the year, covering everything from the basics of machine learning to innovative ideas underpinning computer vision, natural language processing, reinforcement learning, and transformers.
Through my work on NLP models, I started collaborating with Microsoft in regards to generating synthetic text data. We explored the ability of the GPT-3 model to generate both long texts and small extracts containing specific linguistic features. We examined the best approaches for generating prompts, the usefulness of the data thus generated, and the overall impact of text generation on NLP.