Abstract Captis is an exploration of how highly contextualized and individualized LLMs could redefine the field of tracking apps. Based on the real needs of individuals with histamine intolerance and the associated medical therapy phases —elimination, provocation, and stabilization—we developed a functional AI healthcare prototype as a GPT agent, along with a holistic design concept in the form of a conversational UI.
TimeframeÂ
03/2024 - 07/2024
ParticipantsLea Hammann
Andre KlingMethodsUser Survey, Persona, User Interviews, Customer Journey Mapping, Scenario Building, Prototyping, User TestingToolsFigma, Miro, OpenAI, JavaScript, Protopie
Problem descriptionHistamine intolerance (HIT) presents a condition that, while causing significant distress, can be effectively managed through discipline and careful self-planning. This characteristic made the subject particularly interesting, as it offered practical solutions and everyday strategies that could be highly beneficial for those affected. HIT can cause different symptoms in each individual, with varying time intervals between food intake and allergic reactions, and a wide range of food tolerances. These differing variables make a highly individualized and flexible documentation of consumed foods essential to achieve optimal health outcomes. Those affected reported to us that no such app currently exists.
Use Case 1 - Food trackingFood tracking is seamlessly integrated into the conversational chat interface, ensuring a consistent and intuitive user experience. Users can log meals by uploading a photo, which the contextual AI analyzes, while additional details such as ingredients can be confirmed or refined through natural language interaction.Â
This feature addresses the key requirement of reducing user effort, derived from the insight that those with histamine intolerance need simple, quick solutions. The AI-powered database evolves with the user, making meal tracking not only efficient but increasingly personalized over time.
Use Case 2 - Symptom trackingSymptom tracking is another core function within the unified chat interface, allowing users to log and track symptoms as part of the same conversational flow. With minimal effort, users can document symptom type, severity, and progression, while receiving contextual insights based on their previous interactions. The conversational design ensures that the process feels natural and non-intrusive, encouraging users to stay engaged over time. Trends are highlighted visually, bridging the connection between meals and symptoms to provide meaningful, actionable insights.
Use Case 3 - Document generationWithin the same chat interface, users can generate personalized documents summarizing their tracked data, such as dietary patterns, symptoms, and overall trends. These PDFs are designed for practical use, whether for consultations with healthcare professionals or sharing shopping lists with family. By integrating document creation into the conversational flow, the process remains simple and intuitive. This feature is vital for reducing organizational burden and transforming user data into clear, actionable outputs.
The SupergraphicThe Supergraphic, dynamically generated within the chat interface, provides users with a clear, visual summary of their dietary tolerances and health progress. This interactive graphic evolves as users log meals and symptoms, offering a visually compelling way to understand patterns and trends. Designed to be both functional and engaging, the Supergraphic supports users in making informed decisions and can be easily shared with others to communicate dietary needs and preferences effectively. Its integration into the chat ensures accessibility and a seamless user experience.Â
Proof of ConceptAs a proof of concept, we decided to demonstrate the feasibility of our idea by developing a functional prototype. To do this, we chose GPT-4.o as the Large Language Model (LLM) from OpenAI due to its comprehensive features, such as the ability to process and analyze audio, video, and image files. This capability enabled us to analyze food items using these data formats. Since this model is publicly accessible and can be tested by anyone, we used it as a foundation for our experiments. In the initial test runs, we explored the potential of LLMs in image analysis for our use case. Moving forward, we planned to connect the API to a design prototype using ProtoPie. However, we quickly encountered limitations and decided instead to develop our own chatbot within the ChatGPT framework. To build a custom chatbot with clear objectives, a dedicated knowledge base, specific features (such as generating reports or creating shareable lists), and constraints on user queries, we first had to delve into the concept of prompt engineering.
Design processThe design process for Captis followed a user-centered approach, grounded in both qualitative and quantitative research. Through surveys and interviews, we identified the daily challenges faced by individuals with histamine intolerance, such as the complexity of food preparation, difficulties with symptom tracking, and the social and emotional burden of their condition. A key insight was the need for simplicity and personalization—usersÂ
consistently expressed frustration with existingtools that were either too generic or required excessive effort to use. These findings shaped our decision to create a unified conversational interface that integrates all functionalities seamlessly. By focusing on reducing user effort and enhancing accessibility, we prioritized intuitive interactions, contextual AI support, and personalized feedback, ensuring that the app adapts to the unique needs of each user.Â