Speech-Based Alzheimer AI Screening Tool

• Difficulty: Advanced (healthcare and NLP)
• Term Length: Fall 2025 (with possibility of extension to full year)

Description

This project aims to develop a web or mobile app that allows users to record short speech samples. The backend will process the speech with a fine-tuned BERT-family model to estimate Alzheimer’s prediction scores - this is not a diagnosis. The fine-tuned BERT model is based on the project lead’s published research. The tool emphasizes accessibility, affordability, and ethical AI usage.

Skills Gained

• Experience with HuggingFace, PyTorch, and BERT models
• Backend development with Flask or FastAPI
• Web or mobile frontend using React or Streamlit

Technical Details

• ML & Backend: Python
• NLP: HuggingFace + PyTorch
• Frontend: React/Streamlit
• Hyperparameter Tuning: Optuna