Automating product descriptions using generative AI

Solution Overview

In this solution, you will learn how to build a generative AI-powered architecture that processes images using advanced vision models, generating accurate and compelling descriptions of an image. MongoDB Atlas serves as the operational data layer, leveraging its flexible data model to scale seamlessly as new descriptions are added and ensuring efficient data management and effortless scalability.

This solution proves particularly valuable in the retail industry. The process of onboarding a new product to a retailer’s catalog can be time-consuming, especially when it comes to crafting product descriptions—in multiple languages and in different geographies. This solution helps streamline that workflow by automating the initial content creation through generative AI, providing retailers with a solid foundation for each product description, ultimately speeding up the time to market and improving consistency across their catalog UX writing.

The importance of a good description

Did you know that around 70% of people leave a product page when the product description is poor or incomplete?

Product descriptions play a crucial role in the customer's journey. Shoppers rely on them to make buying decisions. When descriptions are weak or missing, businesses risk losing potential revenue and leaving customers frustrated.

Having a great description can enhance user engagement and satisfaction as 87% of online shoppers consider product descriptions to be crucial when making a buying decision. On the other hand, having inaccurate descriptions can result in revenue loss and trust decline with your customer base as 40% of consumers have returned online purchases due to poor product content.

Challenges with the traditional product description writing process

The journey of writing high-quality descriptions involves numerous details and careful considerations. Some of the challenges that present include:

• Crafting compelling descriptions. A product description that is well-written will address their target’s needs and desires while maintaining a consistent tone that aligns with the business’s UX writing strategy and brand identity.

• SEO optimization. Creating SEO-optimized product descriptions to drive organic traffic and improve search engine rankings.

• Multilingual complexity. For retailers with multilingual portals and with multiple operating geographies this process increases in complexity.

• Content approval delays. Even after a description is written, too often there is still a writing approval process that needs to happen, delaying the time to market.

Reference Architecture

At the core of this architecture we have three key components:

• MongoDB Atlas: A powerful, general-purpose data platform that seamlessly manages our data in the cloud. MongoDB's flexible document model allows products to scale effortlessly by adding more descriptions (i.e. for translations in multiple languages) without introducing complexity.

• The object storage: An effective system to save the product image file that ensures efficient retrieval. This can be Google Cloud storage, Azure Blob Storage or AWS S3.

• Together AI: Offers a variety of generative AI services, making it easy to run or fine-tune leading open source models with only a few lines of code. This solution uses their available vision LLMs to generate the product descriptions.

The journey starts on the left side of the diagram with the User/Event label. A laptop icon represents the “Product Description Generator” system. First, a new product is received, this can be either manually by a User or in bulk through an automated Event.

Then, retailers generate the descriptions by sending a Query to Together AI’s endpoint, utilizing their vision models, combining computer vision and natural language processing (NLP) to process and understand images alongside text. The request includes the image URL, the desired description length, the vision model utilized, and the languages for the description.

Then, Together AI takes this data and uses one of its Llama3 vision models to scan the image and generate a description that matches the specified requirements, and returns the Product description to the application.

Finally, the product along with its description will be upserted inside the catalog on MongoDB, ensuring real-time availability across all connected systems.

To illustrate the scalability of this architecture, consider the diagram below. By integrating MongoDB Change Streams, it allows real-time updates on any application listening to the catalog. For example, the e-commerce portal, a social media platform, and any other touchpoints.

Building the Solution

Building this solution can be broken down in five major steps:

Step 1: Replicate the demo database. Provision a cluster within your Atlas account and populate your database with the data required for the demo. A data dump can be found inside the repository to quickly replicate the database with all the necessary data and metadata with one quick mongorestore command.

Step 2: Retrieve MongoDB Atlas connection string. Navigate to your cluster and click on Connect. Copy the connection string provided and save it for your .env file.

Step 3: Create your Together AI account. Sign in to Together AI. Navigate to your account and retrieve your user key, which can be found inside your Profile, then go to Settings and select API Keys. Save this key, as you will need it in your .env file.

Step 4: Create your object storage. Create a new bucket in your AWS account and generate an IAM user with programmatic access. Save the access key and secret key for your .env file.

Step 5: Configure your application’s frontend. Obtain the demo code by cloning the GitHub repository to your local machine, configure the environment variables and install the dependencies. Finally, run the app locally at http://localhost:3003

For complete implementation details, including code samples, configuration files, and tutorial videos, visit the GitHub repository.

Key Learnings

• The importance of high-quality product descriptions. A well-crafted product description enhances user engagement, improves SEO rankings leading to increased views, and helps customers make more informed purchase decisions.

• Leveraging MongoDB and Together AI. By combining MongoDB’s flexible and scalable database with Together AI’s available vision models, retailers are able to automate real-time product description generation that aligns with their business needs.

• The value of modernizing architecture. Streamlining the product onboarding process with AI and automation reduces manual effort and speeds up approvals. Providing a generative AI product description as a base that respects the UX writing and tone of the business ensures consistency across platforms while also enabling faster scalability to accommodate growing product catalogs with minimal additional effort.

Each of these capabilities contributes to building a robust generative AI system that can scale with your business needs while maintaining efficiency and reliability.

Technologies and Products Used

MongoDB modern database

• MongoDB Atlas

Partner technologies

• Together AI
• NextJS

Authors

• Angie Guemes, MongoDB
• Prashant Juttukonda, MongoDB