h2oGPTe Agents: From Model Training to a Live App
- Agentic AI
- Enterprise h2oGPTe
- H2O Driverless AI
This lesson walks through how Enterprise h2oGPTe combines predictive and generative AI through an agentic workflow. It shows an agent trained to predict a target column on the Boston Housing dataset using function calling and generated Python code, then a similar request routed to H2O Driverless AI to train a fraud model, complete with performance graphics and a feature engineering workflow diagram. The lesson also shows the agent writing scoring code for a deployed model endpoint, generating synthetic data to test it, and finally building a working Streamlit application to serve the model's predictions.
Key takeaways
- How an h2oGPTe agent uses function calling and generated Python code to train a predictive model from a prompt
- How the same agent can route a request to H2O Driverless AI to build and evaluate a fraud model
- How the agent generates synthetic data and scoring code to test a deployed model endpoint
- What happens when you ask the agent to wrap a scored model in a Streamlit app: it writes and runs the code
Start owning your AI