Louis Garrett: My summer in Germany was my ticket to the table!
I was welcomed into Prof. Dr. Michael Guckert’s Lab at THM this summer tasked with building an expert Chatbot in rare diseases. Alongside building the application, I was able to enjoy a Summer School and participate in active research.
I am so grateful for the opportunity, which taught me so many skills beyond simple coding that are necessary for a future career in research and NLP.
RareNavigator aims to be a chatbot specialised in replying to natural language prompts about Rare diseases. By the third week I had a created a chatbot capable of traversing its database to find an answer. In testing, it was able to answer prompts accurately maybe, 60% of the time. For a quick prototype, I was quite happy with it.
After understanding the database technology we would be using, I set about finding the data necessary to create an informed chatbot. My supervisor emphasised ‘context’ for building this RAG chatbot and that stuck with me. Obtaining ‘context’ (data) is an integral step in building a chatbot.
Along the way, not all the code I wrote was totally useful. I collected, cleaned and imported textual data from the 1970s, this was not useful. In the final product however, the database was ultimately able to map together diseases, definitions, recent scientific journal articles and authors and their affiliations. Each of these has metadata associated with it including: year published, titles and identification numbers. This resulted in a chatbot capable of providing accurate, science-backed information to professionals and individuals alike.
