E-commerce product fine-tuning with Bright Data and OpenAI

This workflow contains community nodes that are only compatible with the self-hosted version of n8n.

This workflow automates the process of scraping product data from e-commerce websites and using it to fine-tune a custom OpenAI GPT model for generating high-quality marketing copy and product descriptions.

Main Use Cases

• Fine-tune OpenAI models with real product data from hundreds of supported e-commerce websites for marketing content generation.
• Create custom AI models specialized in writing compelling product descriptions across different industries and platforms.
• Automate the entire pipeline from data collection to model training using Bright Data's extensive scraper library.
• Generate marketing copy using your custom-trained model via an interactive chat interface.

How it works

The workflow operates in two main phases: model training and model usage, organized into these stages:

1. Data Collection & Processing

   • Manually triggered to start the fine-tuning process.
   • Uses Bright Data's web scraper to extract product information from any supported e-commerce platform (Amazon, eBay, Shopify stores, Walmart, Target, and hundreds of other websites).
   • Collects product titles, brands, features, descriptions, ratings, and availability status from your chosen platform.
   • Easily customizable to scrape from different websites by simply changing the dataset configuration and product URLs.

2. Training Data Preparation

   • A Code node processes the scraped product data to create training examples in OpenAI's required JSONL format.
   • For each product, generates a complete training example with:
     • System message defining the AI's role as a marketing assistant.
     • User prompt containing specific product details (title, brand, features, original description snippet).
     • Assistant response providing an ideal marketing description template.
   • Compiles all training examples into a single JSONL file ready for OpenAI fine-tuning.

3. Model Fine-Tuning

   • Uploads the training file to OpenAI using the OpenAI File Upload node.
   • Initiates a fine-tuning job via HTTP Request to OpenAI's fine-tuning API using the GPT-4o-mini model as the base.
   • The fine-tuning process runs on OpenAI's servers to create your custom model.

4. Interactive Chat Interface

   • Provides a chat trigger that allows real-time interaction with your fine-tuned model.
   • An AI Agent node connects to your custom-trained OpenAI model.
   • Users can chat with the model to generate product descriptions, marketing copy, or other content based on the training.

5. Custom Model Integration

   • The OpenAI Chat Model node is configured to use your specific fine-tuned model ID.
   • Delivers responses trained on your product data for consistent, high-quality marketing content.

Summary Flow:

Manual Trigger → Scrape E-commerce Products (Bright Data) → Process & Format Training Data (Code) → Upload Training File (OpenAI) → Start Fine-Tuning Job (HTTP Request) | Parallel: Chat Trigger → AI Agent → Custom Fine-Tuned Model Response

Benefits:

• Fully automated pipeline from raw product data to trained AI model.
• Works with hundreds of different e-commerce websites through Bright Data's extensive scraper library.
• Creates specialized models trained on real e-commerce data for authentic marketing copy across various industries.
• Scalable solution that can be adapted to different product categories, niches, or websites.
• Interactive chat interface for immediate access to your custom-trained model.
• Cost-effective fine-tuning using OpenAI's most efficient model (GPT-4o-mini).
• Easily customizable with different websites, product URLs, training prompts, and model configurations.

Setup Requirements:

• Bright Data API credentials for web scraping (supports hundreds of e-commerce websites).
• OpenAI API key with fine-tuning access.
• Replace placeholder credential IDs and model IDs with your actual values.
• Customize the product URLs list and Bright Data dataset for your specific website and use case.
• The workflow can be adapted for any e-commerce platform supported by Bright Data's scraping infrastructure.