/ Project description
Starting point
People who experience racism lack low-threshold services that combine emotional support with legal orientation.
YOUNA set out to create exactly that, across three areas of need:
- Catching and stabilising (emotional coping)
- Informing and educating (e.g. legal information)
- Referral (to the right counselling services or offers)
The first MVP (2022, rule-based, button navigation) quickly hit its limits: too rigid for people to describe their experiences the way they need to.

Challenge
There was no reference project for this use case – needs and scope had to be developed from scratch.
A rule-based bot could not reflect the depth and urgency of each individual case, and the hybrid approach failed on the amount of training data required. On top of that, a topic without an objective “solved”: unlike in customer service, “really helping a person” cannot be measured as a rate. And empathy in handling distressing content does not arise by itself – it has to be designed deliberately.

The road together
Over several years, we developed YOUNA from the rule-based MVP into a tailored LLM system (live since the end of 2024).
At its core is a self-curated knowledge base used by specialised AI agents to answer questions about racist discrimination sensitively and to offer an empathetic first point of contact after racially motivated incidents. Because classic success metrics do not apply here, we developed our own custom metrics and evaluate conversations with Langfuse – quality that otherwise escapes measurement becomes verifiable (with strong results in fact-checking, help for self-help and relevance, among others).
Outcome
People open up – conversations became longer and deeper.


