Overview
This project forecasts which H&M fashion styles will sell best next season and generates new design concepts for them. It builds a weekly panel of styles from two years of H&M transactions and trains a LightGBM model to forecast top-selling styles over the next 13 weeks. For the top forecast styles, it writes design briefs, generates concept images with SDXL conditioned on real reference photos, and checks each concept against similarity, integrity, and attribute tests plus a human review. The same steps are exposed as sub-agents, two reusable Claude Agent Skills, and an MCP server. Under a 13-week training embargo, the model's predicted top 3 land in the true top 20 0.528 of the time over 12 walk-forward origins. The project includes a demo that runs the backtest harness on synthetic data without needing the dataset or a GPU.
Pricing
Free
Detailed plans have not been confirmed in our catalog. Check the official website for current limits and billing terms.
Visit WebsitePrices and limits may change. Confirm the currency, billing period, seat minimum and usage caps on the official website.
Use Cases
- Teams use this to forecast which fashion styles will sell best in the next season.
- Teams use this to generate new design concept images based on forecasted top styles.
- Teams use this to run backtesting and evaluation of forecasting models on synthetic data without needing the dataset or GPU.
- Teams use this to run tests that verify the pipeline without requiring the H&M dataset.
- Teams use this to execute the full pipeline with the H&M data, including data conversion, feature engineering, and model training.
Features
Forecast next season fashion styles
Build weekly panels from transactions
Generate briefs for forecast styles
Create images for product concepts
Check concepts against similarity criteria
Backtest forecasts using synthetic data
Reviews
No reviews yet. Be the first to share your experience!