Ask instead of searching: Chat agent for the HIDA website
HIDA, the Helmholtz Information & Data Science Academy, brings together data science education and training opportunities across various technical and natural science research fields. Our new chatbot makes this diverse range of offerings more accessible: It answers questions about research topics, training courses, or programs, summarizes the most important information, and provides a direct link to the relevant HIDA webpage.
Conversation as a gateway to information
“What Python courses are currently available?”, “What mobility programmes are on offer in Norway?”, and “Are there any PhD vacancies in the field of machine learning at the moment?”
The programme offered by HIDA, the Helmholtz Information & Data Science Academy, is primarily aimed at researchers and doctoral candidates. Events, training courses, collaboration and exchange programmes are developed in partnership with the Helmholtz Centres, German universities and an international network of partners. The focus is on combining scientific and technological research with data science methods.
The chatbot now provides a new way to access this wide range of content: instead of searching for relevant content using keywords and clicking through various pages, you can ‘chat’ directly with the website. The response summarises the relevant information on the topic in question and provides links to the relevant HIDA pages for further details.
Collecting and formulating using the RAG method
A newly developed hybrid method operates in the background. It combines traditional keyword search with semantic search and uses the content found to formulate an appropriate response with the aid of an AI language model (RAG method, ‘Retrieval Augmented Generation’).
To do this, the chat agent accesses indexed data from the TYPO3 website via an interface. This data is prepared for the connected SOLR search and tagged with keywords such as ‘event’ or ‘job’. This allows it to be located and categorized specifically. Based on the question and the sources found, the agent compiles the information, which the language model then uses to formulate the response.
Clear framework – compliant with data protection regulations
Unlike a chatbot with general web access, the content of the HIDA website thus forms both the foundation and the framework for the responses. If the questions lack a clear connection to HIDA’s services, the chatbot will either ask the user to establish a link to them or refuse to provide an answer.
GDPR-compliant data protection was also taken into account during development: Users’ communications with the chat agent are protected by EU law and are not used to train the AI model.
We developed the chatbot in close collaboration with the client. Together, we identified relevant topics based on the website’s structure and defined how responses should be tailored to the various target groups. The chatbot was then customised and optimised to meet these specific needs. The result is a solution tailored to HIDA that provides access to existing services, making it significantly easier to find the relevant information.