Harmful Thought Detector

[Jaishreebala]

Goal: This model must be able to identify harmful sentences within the journal entry.

Caveat: If user does show harmful intent and they reach a point where they get the inbuilt response back, and they want to continue journalling, what do we do? Do we allow them to continue journalling? Maybe make a new entry or start fresh

If harmful, show result in a popup in the frontend

[MeaganShim]

https://huggingface.co/vibhorag101/roberta-base-suicide-prediction-phr?

[MeaganShim]

To do:

• Integrate into aws sagemaker
• Note this can also be reached through an API end point

[Saarangagarwal]

Testing integration with aws sagemaker is more important than creating the model itself given the problems we faced in the past

[MeaganShim]

Look at replicate and see if they have any models for this

[Jaishreebala]

Integrated hf model with ML Trigger Help and tested independently, but the old code wasn't working - sync with BE on this before closing

[MeaganShim]

Next steps - Identify threshold