Plain-language summary
Walking-support chatbots need to understand where a user is, retrieve relevant evidence, and avoid unsupported advice. PacePal combines geographic context with retrieval-augmented generation so that responses are grounded in relevant information rather than generated from model memory alone.
Research contribution
- Geo-aware retrieval for context-sensitive walking-support dialogue.
- Dense retrieval, reranking, contextual compression, and response generation in one pipeline.
- Safety guardrails and system-level evaluation of response quality and latency.
Publication status
Accepted for presentation at the 1st International Conference on Human-Centric Artificial Intelligence (ICHCAI 2026), to be held on 27–28 May 2026 in Halden, Norway. Accepted for publication in IEEE Xplore subject to revisions, IEEE quality and plagiarism checks, and alignment with the conference scope.
Code and dataset
Public code, paper, and dataset links will be added when they become available.
Recommended citation
Chakraborty, S., Shoaib, M., Minghim, R., and Tabassum, M. (2026). PacePal: A Geo-Aware Retrieval-Augmented Chatbot Agent for Walking Engagement and Well-Being. ICHCAI 2026.
BibTeX
@inproceedings{chakraborty2026pacepal,
title={PacePal: A Geo-Aware Retrieval-Augmented Chatbot Agent for Walking Engagement and Well-Being},
author={Chakraborty, Somyajit and Shoaib, Muhammad and Minghim, Rosane and Tabassum, Mujahid},
booktitle={1st International Conference on Human-Centric Artificial Intelligence},
year={2026}
}