Development of an AI to answer questions about documents
Brief description
On the basis of a document archive, the customer wants to be able to formulate questions using natural language, which are then answered appropriately.
Supplement
For smaller texts, a pre-trained LLM can be used directly, which formulates answers with a given context. Various models are evaluated for this purpose, taking into account the runtime and response quality. Frameworks (LlamaIndex, LangChain) are used for larger texts or multiple documents.
Subject description
These frameworks use a vector database in the background that contains the context of the documents used. This context, together with the user's question, is processed by the LLM and an answer is generated from it. The solution approaches are compared and evaluated based on criteria such as runtime and response quality.