AI Content Strategy App

This project provides a simple Streamlit application for running marketing experiments on SME customer data.

Features

• Upload a CSV file containing customer information.
• Automatically determines cluster counts for the industry column and for the product column inside each industry cluster.
• Performs clustering with K-means and summarises each industry cluster by total amount, average credit score and average tenure, ranking them by total amount.
• Allows you to pick how many of the ranked industry and product clusters to include when configuring the experiment.
• Lets you provide separate keywords for marketing messages and product propositions, specify how many of each to generate, adjust a creative temperature slider and optionally generate suggestions using the Gemma-2B-IT model hosted on Hugging Face (requires setting HF_TOKEN).
• Randomly assigns customers to message variants.

Usage

Install the dependencies and run the Streamlit app:

pip install -r requirements.txt
streamlit run streamlit_app.py

Upload a CSV containing lowercase column names:

• industry
• product
• amount
• credit_score
• tenure

The provided sample_customers.csv includes these columns as well as the optional fields cid, company_name and start_date. Column names are automatically normalised to lowercase after upload.

Follow the on-screen instructions to explore clusters, configure your experiment, provide message and product keywords separately and (optionally) generate AI-powered content suggestions.

Gemma model generation requires internet access and an access token for the model on Hugging Face. Set the HF_TOKEN environment variable with your token before running the app. Generation may take a long time on first run.