Cultomato
An AI chatbot that matches Munich's culture scene to your interests and your free evening.
- Role
- UX Designer, Prompt Engineer & Chatbot Developer
- Team
- 3 students
- Context
- University project — HM Munich
- Tools
- Figma, Python, LlamaIndex, Hugging Face, Gradio
Munich has a huge cultural offer, but it's scattered across dozens of sites. Cultomato is a conversational culture guide: tell it what you like and when you're free, and it suggests events. My first AI product — designed and built by a team of three in our first semester.
The problem
Finding something to do in Munich tonight means checking venue sites, listings and social media, and filtering them by taste and time yourself. The effort is high enough that many people simply don't go.
The idea
Replace the search with a conversation. Users say what they're into and when they have time, and Cultomato answers with fitting events. The friendly tomato-can character gives the product a personality that makes "culture" feel less intimidating.
My role
- Designed the conversational experience and interface in Figma.
- Wrote and iterated on the prompts that shape tone and answer quality.
- Built parts of the chatbot with Python, LlamaIndex for retrieval over event data, and Gradio as the front end, hosted on Hugging Face.
Key decisions
- Retrieval over pure generation. Grounding answers in real event data with LlamaIndex reduced made-up recommendations, the biggest trust risk for a recommender.
- Personality as onboarding. The mascot and playful tone lower the barrier to the first question, which matters more than any feature in a chat product.
What I learned
This was my first time shipping something with an LLM inside. It taught me to think about AI features in terms of failure modes (wrong dates, invented events, unclear answers) and to design guardrails for them from day one.