04 / APPLIED AI PROJECT · 2024 / GPT + PREDICTION API
Connecting a conversation to a predictive model.
A prototype connecting a custom GPT to a Flask-hosted insurance premium model, bringing a conversational interface to a predictive workflow.
- My role
- GPT & ML application development
- Context
- Prototype · February–April 2024
- Toolkit
- Python · Flask · scikit-learn · pandas · GPT actions
- /01ConversationCustom GPT
- /02ActionAPI request
- /03Flask APIModel endpoint
- /04PredictionRegression model
Schematic of the prototype's request-to-prediction flow.
The challenge
A predictive model needs an interface before someone can use it. InsureGPT explores connecting a conversation to a model-backed API.
What I owned
I developed a custom GPT with actions connected to a Flask server hosting an insurance premium model. The project used Python, scikit-learn, and pandas, with RandomForestRegressor and OneHotEncoder in the modeling workflow.
How the pieces connect
The GPT provides the conversational interface, an action connects it to the Flask API, and the hosted model produces a prediction. The diagram illustrates that request-to-prediction path.
The public repository contains the application, prediction code, and training notebook.
The result
The project demonstrates a prototype combining conversational interaction with a predictive API: a connected path from user input to model inference.
A prototype connecting conversational input to a Flask-hosted prediction model through GPT actions.
NEXT / ARIZONA STATE UNIVERSITY