Highlighted Updates
- Build AI Agents That Scale
- Get More Accurate Results from RAG
- Control GenAI Costs
- Generate SQL Queries Using Natural Language
- Validate Time Series Models with Better Metrics
- And a Few More Things
Explore feature updates organized by Dataiku capability:
Generative AI
Explore all Generative AI capabilities in Dataiku
Build AI Agents That Scale
AI Agents have arrived in Dataiku. Data Scientists can now quickly design purpose-built agents in the framework of their choice, or capture repeatable logic as plugins to reuse later. In Dataiku, agents don’t stop at experimentation. Connected to trusted analytics and services, powered by the Dataiku LLM Mesh, your agents become powerful agentic applications that can be deployed in a matter of a few clicks.
As more AI agents are created and deployed, the need for quality and trust will also increase. Tracking behavior, managing costs, and having oversight across a growing portfolio of agents becomes a new day-to-day challenge.
To help address this challenge, agents will be first class citizens in Dataiku Govern, ensuring accelerated enforcement of strict qualification and pre-deployment sign-off for all agents. Risk Managers and AI teams can now benefit from a comprehensive GenAI registry which also integrates fine-tuned LLMs to enable both regulatory readiness and agent scaling. Data Scientists and other AI builders can now better examine the specific behavior of each agent through Visual Trace Explorer, benefit from new code agents extensions such as GraphRag, Corrective RAG and Tools-only agents to accelerate their roadmap.
Check out the knowledge base to learn more:
Get More Accurate Results from RAG
Many retrieval-augmented generation (RAG) implementations struggle with documents such as PDFs that contain embedded tables and diagrams. With the new embed documents recipe, you can now build multi-modal knowledge banks that capture both the text and visual elements of your documents, from tables in a PDF to diagrams in a slide deck.
Retrieval can also be challenging when searching for specialized words or phrases such as domain-specific terminology. To help with this, Dataiku 13.4 enables hybrid search, combining both similarity search and exact keyword matching. This enables more accurate RAG, even when working with specialized vocabulary.
Additionally, to ensure output quality, Dataiku 13.4 also introduces new RAG quality guardrails. These provide real-time evaluation of responses from your augmented LLMs using two key metrics:
- Faithfulness: Ensuring responses accurately reflect the source content
- Response relevancy: Verifying answers directly address the given prompt
If responses don’t meet your defined thresholds for either metric, the system will return either an explicit error or custom message, helping maintain high-quality interactions with your augmented models.
Together, these capabilities enable more accurate and reliable GenAI applications, from chatbots to AI agents.
Check out the knowledge base to learn more:
- CREATE MULTI-MODAL KNOWLEDGE BANKS WITH EMBED RECIPES >
- RETRIEVE MORE RELEVANT DOCUMENTS WITH HYBRID SEARCH >
Control GenAI Costs
Running GenAI at scale can quickly become expensive. Dataiku 13.4 introduces new Cost Guard capabilities, enabling you to set and enforce spending limits across your organization. Define budgets by provider, project, or user group, with automated alerts when usage approaches limits.
For full control, you can even block queries that would exceed your defined thresholds – ensuring your GenAI initiatives stay within budget while maintaining alignment between IT and business objectives.
Check out the docs to learn more:
Data Prep
Explore all Data Prep capabilities in Dataiku
Generate SQL Queries Using Natural Language
Writing SQL for data analysis often requires switching between natural thought processes and technical query syntax. Now you can simply describe what you want to analyze using natural langauge, and Dataiku’s new SQL Assistant will generate the appropriate SQL query that you can use in your SQL notebook.
This latest addition to the family of Dataiku AI assistants lets you focus on what you want to learn from your data rather than how to write the query.
Check out the docs to learn more:
Machine Learning
Explore all Machine Learning capabilities in Dataiku
Validate Time Series Models with Better Metrics
Time series forecasting with Dataiku just got even better. In 13.4, you gain granular control over how you evaluate and validate your forecasting models. Access per-fold metrics to understand performance across different time periods, evaluate models with comprehensive statistical insights, and enjoy more precise control over time ranges used for resampling.
Whether you’re looking at year-over-year comparisons or evaluating model performance using residuals, these new capabilities help you build more reliable forecasts.
Check out the knowledge base to learn more:
And a Few More Things
We’re always looking for ways to improve the Dataiku experience. From streamlined navigation to better monitoring tools, 13.4 brings several additional improvements, including:
- Flow Search & Filter: we’ve made it easier to find any item in your flow
- WebApp Monitoring: Track deployed applications in one place
- Multi-Node Deployment: Leverage load balancing for specific applications
- New Date Type: Manage dates with timezones and stay compatible with storage types
- Custom Model Metrics: Add your own evaluation metrics to the model evaluation store