Box Implements Multimodal Enterprise Agents with Gemini Embeddings 2

The way enterprises manage content is facing its most significant structural change since the era of cloud transition. The existing RAG (Retrieval-Augmented Generation) architecture excelled at text-based retrieval but had limitations in handling the multimodal, spatial, and structural data required by the agent era.

Box and Google Cloud are integrating advanced multimodal capabilities into the Box Agentic Platform based on Gemini Multimodal Embeddings 2. This provides next-generation AI capabilities that go beyond text to understand visual elements and spatial structures within documents.

Key improvements include:

• Preservation of Visual and Spatial Geometry: Maintains the spatial layout of complex elements such as tables and financial matrices to accurately interpret relationships between data.
• Utilization of Visual Modalities: Enables search systems to recognize visual elements within documents, such as technical charts, process flowcharts, and product photos.
• Hybrid File Format Linking: Builds an integrated environment that allows cross-referencing of information across different formats, including PDFs, spreadsheets, and presentations.