Scams usually start with messages. Predominantly SMS/email sent with links.

ML models exist to determine this, but we are prototyping using Gemma2 2b to utilise its natural language understanding to see if a better firewall could be built, one which users can talk to.

It has been surprisingly easy to get started on this project: https://ai.google.dev/edge/mediapipe/solutions/genai/llm_inf...

We want to subsequently build a firewall for Calls, too; but that's a more challenging problem given its dynamic and online / realtime nature.

Hope to integrate these two features in our existing network firewall open source app.