Evaluating RAG performance across multiple vector stores and LLMs? We've added an entire suite of tools for that!
Evaluate RAG Retrieval calculates retrieval metrics (Precision@N, Recall@N, FScore@N, MAP@N, MRR)
for a RAG system using an LLM as a judge.
Generate Answers From Context generates synthetic answers for each question present in the incoming records
using a Large Language Model (LLM) from the given context.
Evaluate RAG Answer Correctness evaluates the factual correctness and similarity of generated answers by
comparing against the ground truth answers in a Retrieval-Augmented Generation (RAG) pipeline.
Evaluate RAG Faithfulness evaluates the factual consistency of the generated answer against the given context
in a Retrieval-Augmented Generation (RAG) system.
OpenAI LLM Service integrates with OpenAI's Chat Completion API. Supports configurable parameters,
including model selection, temperature, top_p, max tokens, and retry behavior.
Anthropic LLM Service integrates with Anthropic's Claude AI models through their Messages API.
Supports configurable parameters, including model selection, temperature, top_p, max tokens, and retry behavior.
Summarize Text summarizes the content of a FlowFile or attribute using any of the supported LLM Service Providers.
Consume Kafka improves performance when running stateless and handles partition reassignment by rolling back offsets.
Pinecone and Milvus processors can handle even larger vector sizes
Parse PDF Document delivers continued improvements to complex PDF document parsing.
Query Pinecone now allows for explicit weighting of sparse and dense vectors with hybrid search.
Delete Milvus adds ability to delete records by filter.
UI Updates
Introduced a Provider icon into the extension dialog which is used when creating a new Processor, Controller Service, Reporting Task, or Parameter Provider.
Improved progress indicator when importing a Flow from the Registry.
EL and Parameter syntax highlighting now supports quoted Parameter names.
Prompt Anthropic AI added support for Anthropic's beta features and custom API versions, along with prompting for text vs. images
Publish Kafka added support for zstd compression, and improved support for round robin partitioning to more evenly balance data across the Kafka
cluster
Put Iceberg Table now supports OAuth2 Token Refresh for Polaris Iceberg Catalog
Query Pinecone allows for a configurable number of results via Expression Language
Improved upgrade experience for Datavolo Cloud Runtimes, especially for large-scale BYOC clusters
Improved cluster autoscaling for high throughput and/or bursty workloads, giving you more capacity when you need it and only scaling down when you don't
Delete SFTP now allows you to efficiently delete remote files without first copying them into your Runtime
Upgraded Content Viewer more seamlessly integrates the experience of managing flows and viewing data in-flight
Added a PDF Viewer for a better in-app experience of inspecting PDFs anywhere in your flow, regardless of whether they've been analyzed by
Parse PDF Document
Improved UX for setting properties with long values on processors
PDF Annotator now shows the annotation container hierarchy alongside the PDF. The tree allows the user to easily visualize the hierarchy and how each element relates to others. The tree nodes can be clicked which will jump to that annotation selected in the PDF. The currently selected annotation is shown and its container id can now be copied.
After dragging a new Input/Output Port or Process Group, the user can now simply provide a name and then press Enter to add the component.
Bug Fix: Editing a sensitive parameter description no longer incorrectly changes the parameter value.
Bug Fix: Removed usage of randomUUID function call which wasn't available in all environments.
Bug Fix: Removed action to center selected component when viewing Stats History if the component is already on screen.
Bug Fix: Fixed the Property Combo Editor to no longer show the option to reference Parameters in the Controller Settings where Parameters are not available.
Bug Fix: Introduced new error handling to properly format error messages when the NiFi instance is unreachable.
Searching for components in your data flow once again shows you search results
The "Logout" link is visible on the UI
Fixed conditions for rendering various actions throughout the UI including new Controller Service, new Parameter Context, Move To Parent Group, Upload Custom NAR (when dragging),
and deleting previously uploaded Local Extensions / Custom NARs
Bring Your Own Code - uploading NARs as a Local Extension provides the ultimate flexibility for your Datavolo Runtime whether you're building with
Python or Java
New PDF Viewer makes it easy to see what Datavolo's Advanced Document Parsing discovered as a section header, paragraph, table, chart, etc.
with bounding boxes and metadata for all detected elements.
Multi-user Organizations are now GA! Inviting others to collaborate on your data flows is the latest enhancement to Datavolo Cloud.