Ground search on unstructured and structured data in Data 360 to use generative AI, improve analytics, and use automation tools.
- Add Context to Chunks with Prepend Fields from Data Model Objects
You can now use prepend fields to add additional metadata that provides context for chunks generated from UDMOs. For example, if your search index configuration includes unstructured data from webcrawler or sitemap connectors, metadata fields such as Labels, Title, or URL can provide additional context to each generated chunk. Prepending these fields improves accuracy of RAG responses in Agentforce. Previously, prepend field chunking was available only for search indexes created from DMOs. - Combine Docling Parsing with LLM-Based Image Processing for Accuracy and Cost-Efficiency
Enhance retrieval augmented generation (RAG) responses with information extracted from images, such as flowcharts or diagrams, by enabling LLM-based image processing. When you select the Docling parser, which brings together multiple open-source models to understand document layout and tables, optionally turn on LLM-based image processing for accurate responses with optimal costs. - Customize Retrievers with Advanced Filter Conditions
Include custom logic, define multiple dynamic filters, and use the Like operator to define filter conditions in Einstein Studio for Retrievers. Retriever filter conditions now support more data types, such as Boolean and Date values. Previously, filter conditions supported only one dynamic filter and were limited to text and number data types. - Fine-Tune, Test, and Optimize Retrievers in Retriever Playground (Generally Available)
Customize, test, and validate retriever responses in the Data 360 Einstein Studio Retriever Playground. A misconfigured retriever can lead to irrelevant output, erode user trust, and increase costs. You can verify response accuracy directly in the playground without making calls to an LLM or performing manual testing in local environments. - Measure Retrieval Augmented Generation (RAG) Performance with Retriever Metrics
Use evaluation metrics in the Einstein Studio Retriever Playground to objectively measure the performance of your RAG application. When you test a retriever in advanced mode, the playground generates a context recall score based on a reference answer.