Extensions - Product Recommendations: Bought Together

We are thrilled to unveil a groundbreaking enhancement with our new release: :rocket: Bought Together Recommendations powered by Generative AI and ML Automation. This feature employs a modern MLOps pipeline stack, acting as the backbone of our Product Recommendations Extension.

‍

The Bought Together training flow utilizes advanced techniques. Extracting gender and category attributes from the product catalogue, we employ OpenAI's text-DaVinci-003 model to generate bought-together categories. Our process involves mapping suggested categories with the catalogue's existing ones using OpenAI Clip, Faiss library, and Inner Product metric. Recommendations are then filtered by price percentiles and sorted based on product colour attributes before being stored in MongoDB.

‍

The structured training pipeline involves components like export-cred, read-op, llm-response-op, and dump-op, ensuring efficient data handling and processing. Our services, including DragonFly, Killer, VertexAI, and Santos, collaborate seamlessly to deliver accurate and timely recommendations.

‍

The system design architecture flow illustrates the end-to-end process, from the Product Recommendation Extension making a request to Santos confirming authentication, triggering ML pipelines, fetching data from GCS bucket, and finally serving recommendations to the extension.

‍

This release represents a paradigm shift in product recommendations, enhancing the user experience and driving engagement. Stay tuned as we continue to innovate and refine the future of e-commerce recommendations.