| KEYWORD |
Personal coach based on generative AI
Reference persons MAURIZIO MORISIO
Research Groups DAUIN - GR-16 - SOFTWARE ENGINEERING GROUP - SOFTENG
Thesis type EXPERIMENTAL
Description The goal of this project is to experiment the use of LLMs to build a recommender system in the domain of fitness and well being.
Inputs to the LLM are data about a person’s: physical activity, sleep, food eaten, stress level.
Apart the food eaten, the other parameters are collected from personal wearable devices (aka Fitbit, Smartwatch).
Expected output from the LLM are daily or weekly recommendations about behaviours (notably food, sleep, physical activity)
to improve fitness and well being, customized for the specific person.
Recommendations are based on scientific literature available (Pubmed and other sources) and personal data.
The steps for the project are:
-evaluation and selection of the most suitable LLM (criteria: training on medical data, privacy, licensing)
-construction of ground truth (set of case studies, individual cases and related recommendations by medical experts)
-selection and customization of evaluation metrics
-experimentation on LLM without fine tuning (prompt engineering, design of prompts to convert personal data in specific recommendations)
-construction of data set for fine tuning (medical literature on aging and well being)
-fine tuning of LLM
-experimentation, second round
The project is in collaboration with University of Paderborn and Amsterdam University Medical Center
Required skills Software development, Generative AI
Deadline 31/05/2025
PROPONI LA TUA CANDIDATURA