A New Way to Navigate Social Services with Everyday Language
Source: https://www.tue.nl/en/news-and-events/news-overview/08-03-2026-a-new-way-to-navigate-social-services-with-everyday-language Parent: https://www.tue.nl/en/research
Share
EngD candidate Sichen Guo explores how LLM powered support can make Dutch social services more accessible for people with limited basic skills.
A New Way to Navigate Social Services with Everyday Language
March 8, 2026
LLM powered match bot Mijke helps people with limited basic skills navigate Dutch social services and shows how referral quality can be monitored responsibly.
[Translate to English:]
On March 6, 2026, EngD researcher Sichen Guo from the Department of Industrial Design defended her work on LLM powered support in Dutch social services. Guo, part of the Design Of Empowering Systems ( DoES) cluster, explored how people with limited basic skills can receive understandable guidance in a system that often expects them to search, compare and interpret complex information independently.
Everyday hurdles
For many residents seeking help with living arrangements, finances, health questions, work issues or social participation, finding the right support can be challenging. Local services are spread across multiple organizations, each with their own procedures and formal communication. These routes assume that people can read long texts, understand institutional language and translate it into concrete next steps. Guo shows that these assumptions create a gap between what social service systems expect and what many people can manage when they need urgent help.
A practical companion
Guo designed and operationalised Mijke, an LLM powered match bot that allows users to describe their situation in everyday, conversational language. Instead of formal intake forms, a person can write short messages similar to those used in daily chats. Mijke then responds with practical guidance about relevant organizations, what they can offer and what a suitable next action could be. The work illustrates how such a system can help people overcome hesitation, uncertainty or confusion that often delay timely help seeking.
Behind the scenes
Developing a responsible match bot required more than a workable prototype. Guo focused on building a maintainable foundation that ensures referral quality can be checked and improved over time. She created a structured local service dataset, a versioned prompt library and a documented flow of conversation patterns. She also explored voice interaction combined with persistent text as an additional accessibility direction for people who prefer to speak rather than type.
Quality matters
Using an iterative process with practitioners, Guo defined five criteria that describe what good referrals should look like: relevance, trustworthiness, accuracy, clarity of language and completeness. These criteria support continuous internal monitoring while keeping human judgment central. To assist the monitoring process, she also explored an LLM as judge tool for faster diagnostics and comparison across versions, without replacing practitioner calibration.
Looking ahead
Guo’s work suggests that responsible LLM powered support in social services requires an improvement loop in which data, interaction design and practitioner feedback reinforce one another. Instead of presenting LLMs as standalone solutions, the research shows how they can be integrated into an accountable workflow that helps individuals access support more easily.
Sichen Guo defended her EngD thesis on March 6, 2026.\ Title of the thesis: Operationalising Mijke the Match Bot: LLM Powered Support for People with Limited Basic Skills in Dutch Social Services.\ Supervisors: Jun Hu, Walter Baets.
Marc Rosmalen