Navigating Regulations With Dataiku’s Governance Capabilities

In the last Dataiku AI Governance webinar series session, Triveni Gandhi, Responsible AI Lead at Dataiku, demonstrated how organizations can prepare for the upcoming AI regulations through responsible AI Governance practices with Dataiku. 

While regulations, such as the EU AI Act, introduce clear risk tiers and obligations, much uncertainty remains about how organizations will comply with these new standards. Triveni emphasized that organizations should not halt AI projects out of fear of noncompliance. Instead, they should turn to proper governance frameworks and platforms, like Dataiku, to find help navigating these uncertainties and prepare for compliance.

Key Capabilities for Responsible AI Governance

Triveni introduced several core Dataiku governance capabilities that assist organizations in aligning with AI regulations:

1. Tailored Workflows: Dataiku allows companies to build workflows that match the complexity of their AI projects. High-risk AI use cases often require more comprehensive governance compared to low-risk projects.
2. Pre-Assessment & Project Qualification: Before starting AI projects, organizations can conduct pre-assessments to ensure alignment with internal governance rules. This step prevents non-compliant projects from entering production.
3. Documentation and Traceability: The platform supports documenting sign-offs and model traceability, helping organizations maintain a centralized registry of models, workflows, and decisions to ensure compliance.
4. LLM Governance: As Generative AI and LLMs become more widespread, Dataiku provides tools to govern LLM connections and projects. This includes managing who can access LLMs, conducting audits of query data, and setting thresholds for model performance and data usage.

Demonstration of Dataiku’s Governance Capabilities

During a hands-on demo, Triveni showed how Dataiku’s “Govern Node” acts as a centralized AI and analytics control tower, where companies can monitor and manage the lifecycle of their AI projects. Key features demonstrated included:

• Governed Projects: Dataiku Govern features a dashboard of ongoing AI projects tied to different business initiatives (marketing, finance, manufacturing), categorized by risk and value on a visual matrix. This helps teams prioritize resources for high-value, low-risk projects. The dashboard clarifies the governance status of each project and the project goals.
• Pre-Build Assessment: In Dataiku, teams can build pre-build assessment pages listing core details, such as who the sponsor and developers are, what the project is about, what the expectations are, etc. Once that information is filled out, the team asks for final approval. This is the first sign-off in the AI Governance process. After the assessment is approved, Dataiku creates the project instance with a tag that indicates the project has completed the pre-build assessment. This ensures that only qualified projects move to production. 
• Model Governance: As models progress, teams can monitor metrics like accuracy, bias, precision, and recall. These performance indicators must meet specified thresholds before a model is deployed.
• External Model Governance: Dataiku also enables governance for models built outside of the platform (e.g., in Databricks), ensuring unified governance across various systems.

LLM and Generative AI Governance

The governance of LLMs and Generative AI models is a clear emerging area of concern. Organizations often use third-party models (such as those from OpenAI or Hugging Face). Dataiku allows users to centralize and govern access to these models. With Dataiku, users can:

• Track who uses LLMs and for what purpose.
• Set rules for allowable and forbidden terms in queries.
• Define thresholds for bias and toxicity detection in AI-generated outputs.

Don’t Fret, We’re Here to Help

Triveni concluded by encouraging organizations to begin preparing for AI regulations by centralizing governance processes, conducting pre-assessments, and implementing clear approval workflows. Aligning governance practices with future regulatory requirements may seem like a complex and scary task, but if done with the right tools, like Dataiku, the process can be completed smoothly and efficiently.