[2025F] Noah Sturgill (PhD)

Designing and Evaluating a Conversational Agent for Early Detection of Dementia

Noah Sturgill

Date: 2025-11-07 / 3:00 - 4:00 PM

Location: White Hall 100

Abstract

This project presents the design and evaluation of an empathetic, voice-interactive conversational agent for the early detection of Alzheimer’s disease and related dementias (ADRD). Built upon prior collaboration between Emory and UCSF, the system leverages large language models (LLMs) to guide semi-structured interviews with patients and informants, eliciting cognitive and behavioral symptoms through natural dialogue. The initial web-based prototype, powered by AWS Bedrock (Claude), Whisper, and Kokoro, demonstrated high sensitivity (~80%) to clinician-detected symptoms and strong user acceptance among older adults. To improve accessibility and scalability, the system was extended to a phone-based deployment using Amazon Connect, Lex, Lambda, and S3, enabling remote pre-visit interviews and outreach to underserved populations. Key technical challenges include managing latency, improving automatic speech recognition (ASR) accuracy, and handling multi-speaker diarization in real time. Current work focuses on integrating AWS Transcribe with streaming pipelines and custom vocabularies to enhance transcription quality and clinical usability. This research underscores the potential of conversational AI to augment dementia screening, improve patient narratives, and streamline clinical workflows.

Link

Presentation