Blog

Tavro Agentic AI Governance Playbook – Ingesting LangGraph AI Hiring Agent & EU High Risk Rating into Collibra

LangGraph graph of the hiring agent flowing from start to check_info to generate_response to end

Agentic AI Governance Playbook

The Tavro team used the Tavro Agents to ingest the metadata from LangGraph AI Agents into Collibra AI Governance. We also used the Tavro Agents to automatically assign an EU AI Act risk rating for each agent as well. This “playbook” automated the Agentic AI Governance process via integrations and agents.

LangGraph Agents

LangGraph, created by LangChain, is an open source AI agent framework designed to build, deploy, and manage complex generative AI agent workflows.

As shown in the screenshot above, we developed a very simple Tavro Hiring Agent to automate the recruitment process. The agent included very simple instructions focused on Python or Java skills.

Hiring agent prompt code that selects candidates whose skills include Python or Java

The screenshot below shows the agent responses at various nodes in LangGraph Studio. Because the applicant was a Python developer, the agent responded that the candidate was selected for further steps.

LangGraph Studio showing the hiring agent run, where a Python developer's resume is selected for further steps

LangGraph Agentic Metadata Ingested into Collibra

LangGraph also includes additional configuration parameters such as model name, temperature, description, top_p, and max_tokens.

LangGraph API JSON for the hiring agent showing model gpt-4o-mini, temperature, top_p, max tokens and version

The Tavro team ingested this metadata into Collibra AI Governance.

Collibra AI Agent asset for the LangGraph hiring agent with ingested description, model, temperature, top P and max tokens

Tavro EU AI Act Risk Classification Agent Automatically Scores the LangGraph Agent and Send the Results to Collibra AI Governance

The Tavro EU AI Act Risk Classification Agent received the agentic metadata to automatically score the agent. The screenshot below showcases the Tavro agent UI in Hugging Face (we called this agent via the Hugging Face API).

Tavro EU AI Act Risk Classifier on Hugging Face with fields for an AI use case name and description

Based on the agentic metadata, Tavro classified the Tavro Hiring Agent as High Risk based on Article 6 of the EU AI Act – Employment, Workers’ Management and Access to Self-Employment.

Collibra showing the hiring agent rated High Risk, with EU AI Act Article 6 Employment and Workers' Management marked Yes

This Tavro Agentic AI Governance Playbook combined integrations with agentic auto-classifications to reduce the need for manual AI Governance interventions in Collibra.

NewerThere is an Arms Race to Gather Data to Train AI Models But Organizations Do Not Recognize That Their Data is Often Being Used for Free
← All posts