Idorsia. Design and development of the first Italian Retrieval-Augmented Generation Chatbot for insomnia

With the unconditional sponsorship of the pharmaceutical company Idorsia Italy, it was possible to design and develop a digital assistant for insomnia, capable of answering a wide variety of questions about the pathology, providing support, and disseminating knowledge. AURORA is the first Italian Retrieval Augmented Generation Chatbot. This technology is a natural language queryable search engine that finds the requested information, reprocesses it, and returns an answer in dialogic form, citing the sources.

The development process seamlessly converged at the intersection of MAIZE competencies, which carried out its technological view, and the ones of a board of medical specialists in sleep disorders, who could deep dive into the main topics treated in the chatbot. 

Initially, we co-designed the identity and the main content covered in the chatbot: 

• the thematic areas of interest to build the knowledge base of the chatbot;
• some of the most frequently asked questions by users, within the topic areas;
• the touchpoints on which to deliver the application;
• the visual identity and name of the engine.

Once the topic content was organized we developed the chatbot based on Retrieval-Augmented Generation technology. Retrieval-Augmented Generation (RAG) is an innovative artificial intelligence technology that combines text generation with information retrieval. A RAG tool integrates the functions of a search engine and an LLM-based conversational assistant: the result is a search engine that can be queried in natural language that finds the required information, reprocesses it, and returns an answer in dialogic form, citing the sources. In this specific case, the RAG components are integrated with the proprietary conversational framework “dialog”, which allows for the inclusion of several additional features:

• follow-up questions;
• chit chat;
• NER (Named Entity Recognition) with cross-referencing to entries in a glossary;
• implementation of rules for managing the dialog flow;
• configuration of a wizard;
• orchestrator managing calls to the LLM, optimizing costs;
• integration with external services via API.

The application was also based on domain prompts that control its behavior and tone of voice, with a focus on pharmacovigilance issues.

AURORA can answer questions and concerns related to the daily life of a person suffering from insomnia, give information about this disorder, or advice on how best to manage it. AURORA can be supportive, learn from the questions it receives, and provide valuable advice on how to maintain the healthiest lifestyle possible. While adept in offering support, AURORA conscientiously refrains from encroaching upon the domain of healthcare, by redirecting users to medical professionals when necessary.

To establish both authority and empathy, we opted for the name AURORA. This choice reflects our capacity for creativity and effective communication, drawing on the metaphor of dawn — the moment just before the sun rises. Just as dawn heralds the beginning of a new day, AURORA symbolizes our commitment to guiding users toward clarity and understanding, particularly in challenging situations.

Our technical expertise, combined with the collaboration of sleep disorder specialists and the support of Idorsia, enabled the creation of a knowledgeable assistant capable of providing certified answers with the right level of empathy, drawing from a comprehensive knowledge base.

In conclusion, AURORA stands ready to empathize with users, recognizing the correlations between insomnia and life’s challenges, such as bereavement or separation. It adeptly balances a tone of understanding with a serious and assertive demeanor. Beyond addressing specific concerns, AURORA can also provide insights on broader topics related to insomnia and, for those who are interested, even share the intriguing backstory behind its name and design.