@Sreekanth thanks for the inputs. Asked the same question and got the reply above.
This product is about 1 week of effort on our end on the AI part so its barely an MVP. How it works is we auto parse the pdfs of the policy documents and label each text section. When a question is asked, we have a custom piece of software that picks up relevant parts of the policy text and then feeds the question and policy text to an LLM with some prompting.
We can tweak it by training better on things like how a top up can differ from a regular policy to get more useful answers
Do check other policies and let us know via the feedback tab! :)