As the operator of over 480 IKEA stores in 31 countries, Ingka Group recognizes the value of innovation in driving the retail sector forward.

It’s a philosophy that engineering teams at Ingka Group apply to various aspects of artificial intelligence (AI), including generative AI (gen AI). Which is why, as the need to accelerate time-to-market for gen AI applications continues to grow, Ingka Group has recognized the need for a dedicated platform.

In particular, the organization needed to enable standardized and collaborative workflows across various verticals and deliver new capabilities at scale to its 31 markets. In doing so, the company wanted to ensure observability, promote trust in new technology, comply with relevant regulations, and stay ahead of the requirements of the EU’s AI Act.

KARAN HONAVAR, Engineering Manager, Ingka Group

Central to Ingka Group’s decision to develop its AI projects on MongoDB is the ability to manage structured and semi-structured data on the developer data platform.

“We use MongoDB for all our vector and retrieval-augmented generation (RAG) implementations and large language model (LLM) metadata,” Honavar said, adding that MongoDB also handles the storage of configurations, interaction logs, and performance metrics. However, one important use case is going to be driving the platform’s growth—acquiring co-worker feedback and using it to enhance the performance and predictive capabilities of AI applications.

“The feedback platform is essentially a way to connect data scientists with retail knowledge,” Honavar continued. “Data scientists aren’t retail experts and retail experts aren’t data scientists. This approach helps people focus on what they are great at and captures retail knowledge from across the world. By capturing all this information within MongoDB and interacting with as many co-workers as possible, we can visualize and use the data to fine-tune the models and make them more accurate.”

By using human input through a MongoDB-based workflow to assess the accuracy or suitability of machine-generated recommendations, Ingka Group is constantly refining its AI processes.

“The human data enables us to train the model quickly to comply with IKEA’s values and ethics,” explained Fernando Dorado Rueda, MLOps/LLMOps Engineer at Ingka Group. “This framework allows us to train the e-learning stream, using quantitative data to generate the model weightings.”

The recommendations are also powering a growing range of RAG-based applications that are enhancing the user experience for millions of IKEA shoppers across the world. Automated text translations are becoming more accurate and human in tone, while product recommendations and responses to customer queries are more exact.

KARAN HONAVAR, Engineering Manager, Ingka Group