Iterate over and fine tune model

[barbarastrasser]

Outcome

The focus of this milestone is to iterate and fine-tune the model with available data in the form of already annotated datasets as well as semantic context input (ontologies, etc.). This is especially relevant to improve the retrieval of the Diagnosis and Assessment Tool entities.

Resources:

• Hard coded annotation categories
• Example files used for testing in annotation tool
• An organization containing annotated datasets

Tasks

Note: the code template for the third and fourth task will get ready in #45

• Use RAG to get levels of diagnosis,description for annotation tool and try to increase the output accuracy. #45
• Create embeddings from already annotated data
• Look up suitable resources e.g. mapping from abbreviation of columns to full diagnosis names #40
• Create embeddings from semantically meaningful data sources (ontology, documents, etc.)
• Choose and store embeddings in a vectorstore
• Reorganize LLM invocation - response of queries for document retrieval of vector stores should be included in the prompt handed over to the LLM
• Validate the annotated data