I'm a PhD student at the Department of Computer Science in University of Florida. I am jointly advised by Kristy Elizabeth Boyer and Bonnie Dorr. My research lies at the intersection of Dialogue Systems, Education, and HCI. I also like to build and contribute to open-source software tools and libraries for ML engineers and researchers. I am affiliated with LearnDialogue as well as the NLP group @ UF.
- 🌱 I’m actively looking for Research Internship openings for Summer 2024.
- 💬 Ask me about anything, I am happy to help. I am very much open to new research ideas and collaborations.
- 📫 You can reach out to me at msamogh@gmail.com (more social links at the bottom).
Commercially-motivated "task-oriented" and "chitchat" dialogue systems such as virtual assistants and customer service bots are all around us and are more powerful than ever thanks to LLMs. However, advances in LLM-based dialogue systems are yet to translate into educational dialogue, negotiation, collaboration, and so on. I collectively term such paradigms as the "Long Tail" of dialogue systems.
My research interests touch upon ML, NLP, and HCI issues surrounding the authorship, bootstrapping, crowdsourcing, synthetic data generation, and evaluation of these Long Tail Dialogue Systems.
- LLM4Qual: LLM4Qual is an open-source framework that lets you combine HuggingFace EvaluationSuite with LangChain's LLM capabilities for qualitative research.
- nonechucks: nonechucks is a library that provides wrappers for PyTorch's datasets, samplers, and transforms to allow for dropping unwanted or invalid samples dynamically.
- Rasa Frames: Most dialogue state trackers only support one value per slot. Rasa Frames is a proof-of-concept fork of Rasa Core that supports multiple frames that allows for multiple values per slot. This is helpful, for example, when the user is comparing two flights.
Rasa | TensorFlow | MSCKF-VIO
