How We Saved Hundreds of Hours for Content Creation by Integrating OpenAI to the Marketplace

Challenges

Previously, we assisted WheelerShip in expanding from a small online store to a multi-vendor marketplace, and currently, we are focused on continually adding more vendors to their system.

Numerous vendors share information about their products through APIs or Excel sheets, which contain comprehensive details about their offerings, including brand, size, style, condition, and product description.

In one instance, we encountered a situation where we needed to integrate a new tire vendor into the WheelerShip platform. However, this particular vendor had up to a thousand of products available and did not provide any product descriptions. Lack of information about products can lead to poor SEO, lower conversions, and ultimately, reduced sales. WheelerShip understands the importance of providing comprehensive product information to avoid these negative outcomes and safeguard their business.

Consequently, we encountered the need to efficiently fill in the product descriptions without consuming excessive time. Thus, we sought a swift and precise solution to obtain the necessary data.

Discovery Stage

All the vendor data was successfully delivered to Magento through Data Flow Application (DFA), specifically customized for WheelerShip. DFA functions as middleware, facilitating seamless communication between multiple systems, particularly Magento and the vendors.

The communication between these systems is facilitated by adapters. In this case, an application programming interface (API)* adapter was utilized to transfer the data from the vendors since they provided the information via APIs. Subsequently, the data was transmitted to the brain of the application, which operates as the central hub responsible for routing requests among the different systems. The application’s brain possesses the necessary knowledge regarding the specific data requirements and destinations, allowing it to effectively coordinate the flow of information between the systems involved. Finally, the eCom adapter transmitted the data to Magento for further processing and integration.

So it worked like this:

However, WheelerShip faced a challenge as they did not have product descriptions provided by this specific vendor. The primary concern was the prospect of hiring a team of content writers and managers, which would be both time-consuming and costly. So here comes the brainstorming of alternative approaches to avoid such routine tasks.

Given the recent buzz surrounding OpenAI*, we immediately turned our attention to exploring its capabilities. OpenAI offers APIs to provide developers with powerful tools to integrate natural language processing capabilities into their applications. These APIs leverage cutting-edge deep learning models, such as text-davinci-003**, to understand and generate human-like text responses.

With the text-davinci-003 model, developers can tap into an advanced language model that excels in instruction-following tasks. It is specifically designed to respond concisely and accurately. This model represents a significant advancement in generating high-quality, context-aware text responses, making it ideal for content generation.

So here it was, the solution that should be a perfect fit for our case!

Development

To integrate OpenAI with WheelerShip, we established a connection between OpenAI and DFA using a custom API adapter designed specifically for system communication. The text-davinci-003 model was designated as the target for the requests. Thus, when some product of the tire vendor lacked description, DFA would send a request to OpenAI via APIs and text-davinci-003 generated it.

As WheelerShip deals with multiple vendors and a wide range of products, we needed to configure OpenAI requests specifically for generating product descriptions for the tire vendor. Accuracy and SEO-friendliness of the received data were also crucial considerations.

We configured DFA to send requests to OpenAI based on tire brand and name, and in return, receive a relevant product description. Since WheelerShip had limited space for descriptions, we set a character limit of 100 symbols to ensure concise content.

However, we encountered a challenge with OpenAI’s text generation. It sometimes produced text longer than the specified 100 symbols, while DFA only accepted inputs of that length. Yet, text-davinci-003 could not be asked to generate text in any other way. To address this, our developers came up with a backend solution to handle this aspect effectively. They programmed DFA to receive the text only after a full stop.

Another issue arose when OpenAI occasionally generated irrelevant content such as pieces of code or unrelated links. To mitigate this, our developers configured the backend side to prevent such data from entering DFA, ensuring the generated information remained relevant and appropriate for the product descriptions.