In 2024 we developed Lupai, an AI-powered assistant designed to help migrants navigate labor and migration rights in Germany. The goal was not to replace existing advisory services, but to make reliable information more accessible to people who often face complex bureaucratic systems, language barriers, and fragmented sources of information.
The idea behind Lupai was inspired by a familiar practice within migrant communities: mutual support. People who have already gone through administrative processes often help newcomers understand how to deal with visas, employment contracts, or official letters. These informal networks of solidarity are essential—but they also have limits. Not everyone has access to such networks, and advisory organizations often face more demand than they can meet.
Could an AI system help extend access to this knowledge while respecting the importance of human counseling and community support?
Over the course of the project we encountered a number of challenges that go beyond typical software development: how to curate trustworthy knowledge, how to integrate experiential knowledge from communities, how to design AI systems that handle sensitive topics responsibly, and how to deal with contradictory information in official sources.
In a new paper, we share the lessons learned from building Lupai. The paper discusses:
Our hope is that these insights can be useful for others working on AI for social good, especially in areas where technology intersects with rights, public services, and vulnerable communities.
You can read the full paper here