/docs-archive/old/forms/reports/q3-2021/ · last edited 01 Jul 2018 · Operations
For the Polygon platform, we decided to implement a local AI assistant, primarily for utility and entertainment purposes. Given our target audience—family and friends—using a popular pre-trained model like Claude or Turing wasn't feasible due to licensing and computational resource constraints. Instead, we opted for an open-source, lightweight model called AssistantNet. This choice balances the need for functionality with our limited resources. AssistantNet is adept at providing basic information, answering questions, and assisting with scheduling—tasks that are sufficient for our needs while keeping the system lightweight and responsive.
Moreover, local hosting of the AI assistant ensures better privacy and security, as we don't have to rely on external servers. This is particularly important given that the platform will be used by a close-knit group of individuals who might be sensitive to data handling practices. The AI assistant can be periodically updated through our internal server, ensuring it stays relevant without the need for frequent external interactions.