Evolutionary AI Chatbots

Evolution means to dynamically change over time, according to the expectations that's required to facilitate for the survival of the fittest. Speaking about evolution doesn't make sense in relationship to AI, but it's difficult to not use lofty words as we're describing our newest feature. Basically, as you're having a conversation with your AI chatbot, you can modify its behaviour as you see fit. Watch the following video to understand what I'm talking about.

CRUD on your RAG database

In the above video I'm having a conversation with an example AI chatbot. The chatbot is connected to the internet and can search the web, in addition to dynamically updating its own internal RAG database during the conversation according to your instructions.

The feature came up because one of our clients are performing online investigations of persons of interest. I can't disclose who, but as he is performing his investigation often times the database becomes "polluted" with irrelevant content. Regardless of who he's conducting a search for, there's always a soccer player somewhere with the same name.

To increase the quality of his data, he needed the ability to dynamically modify the RAG database during usage, and delete items, in addition to performing standard SQL-based lookups into his RAG data.

Both of these were easy for us to solve due to our AI functions, and basically just implied implementing CRUD capabilities on the underlying training material. To understand, imagine the following training snippet being a simple training snippet added to the model as a training snippet record.

The above training snippet becomes an instruction to the model, allowing us to say stuff such as "Delete the training snippet with the ID of xyz" for instance.