Liminal

The Journey

During spring break, my team and I developed an AI guardrail system designed to detect mental health crisis in youth interacting with our AI model. Initially, we trained the model on the provided seed dataset; however, we soon realized it was overfitting, as evidenced by perfect 1.000 recall and precision scores. To address this, we engineered a synthetic data pipeline that stress-tested our ungoverned model, successfully diversifying the training data and resolving the overfitting issues.This project taught me a great deal, as it was my first time working with an AI model as a developer. Through this experience, I learned the fundamentals of how to train and refine an AI model, which is an important asset these days.