Inspiration

In 2023 alone, there have been 46,563 reported scam cases - a 50% rise from 2022 - and more than $650 million has been lost to such scams. By far the fastest rising group of scams so far are "fake friend" call scams, which have risen 225% in the past year.

There are not many ways to counter such scams - rule-based and blacklist-based softwares such as ScamShield are unable to keep up with the vast amounts of new phone numbers used for scamming, and no existing approach is able to filter the call by what the scammer says to block the call in this manner to block such a scam.

What it does

FriendZone is a service that allows users to buy a protected phone number.

  • These phone numbers will use an automated platform (Twilio) to make all first-time callers say their name, which will be recorded by speech recognition software.
  • The call is routed for the actual user to pick up and is monitored in real time by AI (powered by GPT).
  • If the AI detects anomalous behaviour by the caller, such as repeatedly asking for the user to guess their name, the call will be flagged and ended automatically.
  • If the AI detects names being used that are different from the initial name given, the call will also be flagged.

How we built it

We used Twilio to control the protected numbers, allowing us to use basic text to speech functions, stream the call to Flask for further processing, as well as dial a real phone number to as to forward the call. On the processing side, we used Vosk to recognize the speech said, then used GPT-4 to parse what was said to determine if the caller is indeed a scammer or not.

Challenges we ran into

Twilio does not offer phone numbers in Singapore, so we had to do international calls to the UK, and it was quite expensive until we figured out how to properly do international calls.

Accomplishments that we're proud of

We're proud of being able to successfully work with phone calls, as it's our first time doing so.

What we learned

How to manage phone calls using Twilio.

What's next for FriendZone

Implementing multilingual speech recognition due to the multilingual nature of scams. We could also make the software work with more types of scams, such as jobs scams, which are also on the rise.

Built With

Share this project:

Updates