A half-day, hands-on tutorial on building and adapting multi-agent LLM systems with real-world sensing data.
Explore the Tutorial ↓Large language models are transforming how researchers analyze and draw insights from passively collected sensor data. Multi-agent LLM systems — where specialized agents collaborate on planning, data retrieval, code execution, and sensemaking — offer a powerful paradigm for ubiquitous computing. Yet adapting these systems to real-world sensing problems introduces unique challenges around data representation, privacy, agent orchestration, and missing data that are rarely discussed in research papers.
This tutorial bridges that gap. You will leave with both the conceptual grounding and the practical skills to begin applying multi-agent systems to your own datasets and research problems.
Understand how specialized LLM agents collaborate — from action planning to code generation and sensemaking.
Tackle real problems: noisy data, missing values, privacy constraints, and longitudinal study design.
Run and modify GLOSS — an open-source multi-agent framework published at IMWUT 2025 — on your own machine.
Walk away able to adapt a multi-agent system to your own dataset and observe how changes affect system behavior.
Introduction to multi-agent LLMs for UbiComp. Get up to speed on the landscape — from prompt engineering and RAG to why multi-agent architectures are gaining traction in sensing research.
Explore GLOSS live. Run the GLOSS multi-agent network on your own device using a provided sensing dataset and interact with it via open-ended natural language queries.
Hear from real applications. Team members share 10-minute demos of how they adapted GLOSS to their own research projects — with live codebase walkthroughs showing actual modifications and their effects.
Complete curated hands-on tasks. Work through guided exercises on the College Experience Dataset — prompt engineering, integrating new sensor streams, and modifying output format constraints — and observe the effects firsthand.
Group discussion. Close with a structured discussion on challenges, limitations, and open research opportunities in multi-agent sensing systems.
The tutorial runs as a single half-day session. Here's the plan:
Overview of LLMs in UbiComp, common approaches (prompt engineering, RAG), and the rise of multi-agent systems in sensing research.
Deep dive into GLOSS architecture, environment setup on participant machines, and 10–15 minutes of free-form exploration with open-ended queries.
Connect with fellow researchers over coffee.
Back-to-back 10-minute presentations from the organizing team on real adaptations of GLOSS — activity annotation, health monitoring, and more — with demo videos.
Guided exercises on the College Experience Dataset: prompt engineering, sensor stream integration, output format constraints. Two to three tasks, 10–15 min each.
Group reflection on challenges, open problems, and next steps for applying multi-agent systems to your own research.
This tutorial is designed for participants with a wide range of backgrounds. No prior experience with LLMs or multi-agent systems is required. Basic Python familiarity and some exposure to wearable sensing data are helpful, but the tutorial is intentionally accessible. Participants without a compatible OS or coding background will be paired with team members to ensure full participation.
Mobile & Wearable Sensing
Core audience — directly relevant to your data pipelines.
Machine Learning
Explore a new paradigm for structured reasoning over sensor data.
Behavioral & Health Science
Understand how AI can turn passive data into actionable insights.
Intervention Design
See how multi-agent systems can power just-in-time adaptive applications.
PhD Students & Early Researchers
Build practical skills you can apply to your own dissertation work immediately.
Curious Attendees
Non-technical backgrounds are welcome — team members will be paired with you throughout.
Organized by researchers from the UbiWell Lab and mHealth Research Group at Northeastern University. Questions? Reach out to the co-leads.
Register for the tutorial through the official UbiComp / ISWC 2026 website. Tutorial materials will be made available for download after the event.
Tutorial codebase, datasets, and slides will be available here after the tutorial ends.
Coming Soon