Large language models (LLMs) excel at natural language operations (translation, grammar correction, information retrieval) since they were trained to discover sentence context, but they usually falter when factual information is required, due to a lack of real understanding of the vast ammount of information in their models. This intrinsic limitation makes them poor tools for medical tasks where factual, relevant, vetted information is paramount.

LLMs are also very expensive in terms of computational resources. This translates in high costs and worse: high response times for most of the interesting use cases. A strategy based only on LLMs does not scale to high volume applications like PGD for prevention and follow-up where the LLM may be presented with full patient history frequently to monitor for changes.

AI is not only LLMs, and there is much to be gained by combining other techniques that have greater speed and higher accuracy at the expense of needing a more structured approach. We combine multiple AI models (DL, ML, generative) to make it predictable, reliable and economically feasible for continuous monitoring.

The system is multiuser and each user can have private libraries with different subsets of the OA Set and/or upload their own documents. Of course community libraries amongst groups of users are also possible. Alerts can be setup when new relevant research to a library is uploaded to the OA Set.

Information already processed, vetted and stored in the knowledge database is used to continuosly enhance the accuracy of knowledge mining of the system and to produce more and better artifacts (summaries, patient recommendations, drug/disease interactions, etc).

This information workflow can be applied to any other knowledge area since multiple knowledge bases can be established where experts are now users and viceversa and even designated for read-only access. Share artifacts with the public or via controlled access.