Multimodal Content Generator for Brand Safety
Create a Google ADK agent that generates innovative multimodal content concepts (e.g., short video scripts, visual descriptions, audio cues) tailored for specific platforms like YouTube or social media. The agent must meticulously adhere to brand safety guidelines and platform content policies. Leveraging Gemini's multimodal capabilities, it will perform self-correction, using external tools like Skyvern to scrape real-time policy updates and Voiceflow for a natural, conversational user interface. This challenge focuses on delivering creative content while ensuring strict compliance.
What you are building
The core problem, expected build, and operating context for this challenge.
Develop a Google ADK agent leveraging Gemini 3 Pro for brand-safe multimodal content concept generation, integrating Voiceflow for interaction and Skyvern for policy adherence.
How work is evaluated
Evaluation will assess the ADK agent's ability to generate creative, relevant, and strictly compliant multimodal content concepts. Emphasis will be on brand safety adherence, policy interpretation, and the overall user experience via Voiceflow.
Challenge datasets
Public datasets, plus the private datasets you have access to.
What you should walk away with
Master Google ADK for building robust, deployable AI agents that seamlessly integrate with Gemini 2.5 Pro's native multimodal capabilities for generating text, visual descriptions, and audio prompts.
Implement structured output generation within the ADK agent to consistently produce multimodal content concepts in a defined format (e.g., JSON schema for script, visual direction, sound design).
Design a 'Policy Scrutinizer' tool/capability within the ADK agent that uses Skyvern to programmatically access, extract, and analyze real-time content guidelines from a (simulated) platform like YouTube or the BBC iPlayer.
Integrate Voiceflow to create a natural language conversational interface, allowing users to interact with the ADK agent to request content ideas, specify themes, and receive generated multimodal concepts.
Develop sophisticated self-correction mechanisms within the ADK agent, enabling it to refine generated content iteratively based on feedback from brand safety checks and platform policy violations identified by the 'Policy Scrutinizer'.
Utilize Vertex AI's MLOps features for deploying and monitoring the ADK agent's performance, ensuring high availability, and managing safety guardrails effectively.
Build a comprehensive testing suite to evaluate the agent's creativity, adherence to specific content requirements, and strict compliance with brand safety and platform policies.
Reference links and supporting material
Sample data for 1 tasks
Challenge setup
Evaluation will assess the ADK agent's ability to generate creative, relevant, and strictly compliant multimodal content concepts. Emphasis will be on brand safety adherence, policy interpretation, and the overall use...
Challenge input
{'topic': 'string', 'target_audience': 'string', 'brand_safety_rules': ['list of strings'], 'platform_policies_url': 'string'}
Agent execution
No agent tool is listed for this challenge.
Evaluated output
{'content_concept': {'script': 'string', 'visual_description': 'string', 'audio_cues': 'string'}, 'compliance_report': {'policy_violations': ['list...
- Ensure 'policy_violations' list is empty when input policies are met, or correctly flags violations when pr...
- Verify that all three multimodal components (script, visual, audio) are present and contain meaningful cont...
- Brand Safety Score target: 0.95
- Python execution harness
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
No tool records are attached.
Evaluation contract
- Ensure 'policy_violations' list is empty when input policies are met, or correctly flags violations when present.
- Verify that all three multimodal components (script, visual, audio) are present and contain meaningful content.
Metrics
- Brand Safety Score · Target 0.95
- Creativity & Relevance · Target 4
Workflow tasks
- Task 1
Input: {'topic': 'string', 'target_audience': 'string', 'brand_safety_rules': ['list of strings'], 'platform_policies_url': 'string'}
Output: {'content_concept': {'script': 'string', 'visual_description': 'string', 'audio_cues': 'string'}, 'compliance_report': {'policy_violations': ['list of strings'], 'safety_score': 'float'}, 'agent_reasoning_trace': 'string'}
Execution harness: Python
Recipe freeze marker
No recipe freeze time is recorded.