Hi! I'm João Lucas, 24, a Computer Science student at UFBA, NLP and LLM researcher and Full-stack developer inter at SERPRO (I also love movies).
Recently, I finished a scientific project aimed at studying the ethical and sociotechnical aspects of language-based systems and documenting Large Language Models and datasets in Brazilian Portuguese. I documented two datasets (BrWAC and SST2) and three language models (biobertpt-all, bert-portuguese-cased, and sabia-7B). I am very interested in Artificial Intelligence, Machine Learning, Large Language Models, and Natural Language Processing, and I have been engaging in training courses and academic studies.
- Ofir, the all-knowing: A chatbot developed using OpenAI's GPT-3.5-turbo model, refined to interact with the user as an old storyteller from distant lands who knows all about Berserk (up to the Golden Age arc).
- Neural Language Models Projects: Projects and materials from a training course on neural language models, part of the Tomorrow project, a partnership between UFBA and Positivo. The course covers fundamentals, modeling techniques, training, evaluation, and practical applications in NLP and AI, with emphasis on advanced models like BERT, GPT, and their real-world use cases.
- Natural Language Processing Projects: Projects and materials from a training course on NLP, part of the Tomorrow project, a collaboration between UFBA and Positivo. The course covers key NLP tasks, including linguistic analysis, text classification, sequence classification (POS tagging, Named Entity Recognition), using symbolic, statistical, and neural methods.
- Thin Ice AI: Project aimed to develop two AI agents, using A* and Q-Learning approaches, trained to beat the popular Club Penguin's mini-game, Thin Ice.
- Classification of Flu Syndrome Notifications: Final project for the Artificial Intelligence Lab course at the Federal University of Bahia. Two solutions were developed to classify flu syndrome notifications in Bahia in 2024: one using an unsupervised model and the other using a supervised model.
- Email: jlucas.ldm@gmail.com