In the fast-paced world of e-commerce, innovation is key to staying ahead. We recently partnered with an online retailer to transform their search functionality, creating a bespoke visual search solution that sets them apart from the competition.

The Challenge

Our client observed a significant trend: customers were increasingly seeking products inspired by images from platforms like Pinterest. Traditional text-based search wasn't cutting it—customers often lacked the precise keywords to find what they were looking for. The result? Frustration and missed sales opportunities.

Our Custom-Built Solution

Rather than relying on third-party API integrations, we took on the challenge of building a tailored visual search system from scratch. Here's how we did it:

1. Wrapping the Neural Network in a Custom Container

   We utilized the VGG (Visual Geometry Group) neural network, a powerful deep learning model renowned for image recognition. To ensure seamless integration and full control, we wrapped this neural network within our own custom container. This approach allowed us to fine-tune the model to the client's specific needs and maintain flexibility for future enhancements.

2. Deploying as an AWS Lambda Function

   To make the solution scalable and cost-effective, we deployed the neural network as an AWS Lambda function. We maximized the allocated RAM to push the performance to its limits—operating right on the edge of the requirements. This serverless architecture ensures that the system can handle varying loads without the overhead of managing servers.

3. Creating a High-Performance Vector Database

   By processing the client's product images through the VGG network, we converted each image into a high-dimensional vector that captures its essential features. These vectors were stored in a database optimized for rapid similarity searches.

4. Enabling User Image Queries

   When a user uploads an image (whether a Pinterest screenshot or a personal photo) we process it through the same neural network container to obtain its vector representation. We then compare this vector against our database to find visually similar products.

The Results

Our bespoke solution is predicted to deliver outstanding outcomes:

• Superior User Experience: Customers can effortlessly find products matching their visual inspiration without needing exact keywords.
• Increased Engagement: The intuitive search encourages users to explore more, boosting engagement metrics.
• Sales Growth: Simplifying the search process led to higher conversion rates and increased average order values.
• Full Control and Flexibility: Owning the solution end-to-end allows for ongoing customization and scalability without dependency on external APIs. This allows us to host your application in region of your choice and ensure your customer data never leaves EU.

Why Build Your Own Solution?

By developing a custom container and deploying it as a Lambda function, we achieved a level of performance and control that third-party APIs couldn't offer. This approach ensures data privacy, reduces long-term costs associated with API usage, and allows for tailored optimizations specific to the client's needs.

An Open Question

Building a custom solution requires an upfront investment but offers unmatched flexibility and control. On the other hand, integrating with existing APIs can accelerate deployment but may limit customization and entail ongoing costs.

What Would You Choose?

Is developing a bespoke solution the right path for your business, or would integrating with existing APIs suffice? We'd love to hear your thoughts and discuss how we can help you achieve your goals.

Get in Touch

Ready to explore how a custom-built visual search can elevate your e-commerce platform? Contact us to start the conversation.

---

At CherryPeak, we specialize in crafting innovative, tailor-made solutions that leverage the power of AI and machine learning to drive business success.