As part of the Multimodal User Interfaces course, our group developed ParkDas, a Telegram-based parking assistant for the University of Ulm campus. The project focused on designing a multimodal assistant that helps users find suitable parking options quickly and with minimal distraction. A particular emphasis was placed on natural interaction, map-based output, and practical usability in a real campus setting.
Overview
Finding parking on campus can be time-consuming, especially when parking availability changes frequently and relevant information is not easily accessible while on the move. The goal of ParkDas was to address this problem with a conversational assistant that combines text-based interaction, location-aware recommendations, and visual map output.
The resulting system is a Telegram bot that helps users identify suitable parking areas, navigate toward them, and access current occupancy information more conveniently. Instead of relying on a rigid command structure, the bot supports more natural requests and uses Google Gemini to interpret user intent and extract relevant information such as destinations, time references, and parking preferences.
Interaction Design
A central design goal of the project was to create an interaction flow that remains practical in a driving or walking context. The assistant therefore supports conversational input and returns concise, situation-specific recommendations rather than requiring the user to navigate a complex interface.
To support multimodal interaction, the bot combines natural-language responses with map-based visual output. Parking locations are displayed using labeled and color-coded markers, which makes it easier to compare alternatives and understand their relative position. In addition, the system can incorporate location data to provide more tailored recommendations based on the user’s current context.
This combination of conversational input and visual feedback made the assistant more flexible and more immediately usable than a purely menu-driven approach.
Implementation
The project was built around three main technical components: text understanding with Google Gemini, map generation using Google Static Maps, and a custom backend layer for interaction logic and data management.
On the language side, Gemini was used to interpret user requests beyond simple keyword matching. This included understanding user intent, identifying time and location references, and deriving recommendations from free-form text. This made the bot substantially more flexible than a rule-based command interface and allowed it to respond more naturally to different phrasings.
The backend handled conversation flow, state management, and the storage of user-specific data and settings. The system was also designed with internationalization in mind, allowing support for multiple languages. On the output side, static map images were generated dynamically in order to visualize parking options and improve the overall usability of the recommendations.
Within the project, I was responsible for backend development, interaction logic, and the integration of the AI-supported processing pipeline.
Conclusion
ParkDas was a well-received project that combined multimodal interface design, practical backend development, and a lightweight but meaningful use of AI-driven language understanding.
The project was particularly valuable as an exercise in designing a user-facing assistant that integrates multiple modalities into a coherent workflow. It also demonstrated how large language models can be used effectively in a constrained application setting to improve usability without requiring a fully open-ended conversational system. The source code is available on GitHub.