![Best architecture for building an ai text detection system with deberta-v3 and fast api [closed]](https://stackoverflow.com/Content/Sites/stackoverflow/Img/apple-touch-icon@2.png?v=0f0cab681579)
Want to improve this question? Guide the asker to update the question so it focuses on a single, specific problem. Narrowing the question will help others answer the question concisely. You may edit the question if you feel you can improve it yourself. If edited, the question will be reviewed and might be reopened.
This is the description and phases of the product im going to build- enter image description here
My goal is to build ai malign detection platform that
detects ai generated texts across websites
Checks this if the ai generated texts are malign , malicious , attacks or phishing
If it does , it’ll do provenance tracking of the flagged information
It uses explainable ai (SHAP) to explain why it was flagged
Displays the report in a web dashboard.
For phase 1, I have tried building a browser extension that detects ai texts across social media and news websites.
For that, i have fine tuned a pretrained model (deberta-v3-base) with datasets like defactify , openturing and raid benchmarks.
But im not getting a good evaluation score of my model since im new to fine tuning.
After traning , while testing , my model predicts everything as ai generated texts across all websites. I realised that my model has become biased
I need to complete phase of this project as soon as possible so your help would be appreciated.please help me