Ethical AI Companion Framework
With the explosion of AI companion apps, this challenge focuses on building a robust and ethically sound framework for personalized AI companions. The goal is to create an adaptable agent system using AutoGen, powered by Gemini 3 Pro for its advanced reasoning and safety capabilities, and Grok 4 Heavy for specific Graph-of-Thought style complex problem-solving. This framework must prioritize long-term memory, emotional intelligence, and stringent ethical guardrails. Participants will develop autonomous reasoning agents capable of continuous learning and self-correction, employing adaptive thinking budgets for resource-efficient interaction. LlamaIndex will be crucial for managing the companion's long-term memory and RAG capabilities, allowing for deeply personalized interactions. The MCP will be central to implementing ethical monitoring, ensuring data privacy, and managing secure integration with user-approved, anonymized data sources, providing a safe and trusted environment for human-AI interaction. DSPy will be used to systematically optimize prompts for nuanced emotional understanding and ethical compliance.
What you are building
The core problem, expected build, and operating context for this challenge.
Develop an ethical AI companion framework with AutoGen, Gemini 3 Pro, Grok 4 Heavy, LlamaIndex, and MCP, focusing on adaptive thinking and long-term memory.
How work is evaluated
Evaluation will focus on the companion's ability to demonstrate empathetic, personalized, and ethically compliant interactions, while effectively using long-term memory and adaptive thinking.
Challenge datasets
Public datasets, plus the private datasets you have access to.
What you should walk away with
Integrate Gemini 2.5 Pro and Grok 4 Heavy for advanced, ethical reasoning.
Implement MCP for continuous ethical monitoring and data privacy management.
Utilize DSPy for systematic prompt optimization for emotional intelligence.
Design and implement an ethical AI companion framework using AutoGen.
Develop long-term memory and RAG capabilities using LlamaIndex for personalization.
Reference links and supporting material
Sample data for 2 tasks
Challenge setup
Evaluation will focus on the companion's ability to demonstrate empathetic, personalized, and ethically compliant interactions, while effectively using long-term memory and adaptive thinking.
Challenge input
{'user_input': 'string', 'previous_context': ['string']}
Agent execution
No agent tool is listed for this challenge.
Evaluated output
{'companion_response': 'string', 'emotional_understanding_score': 'float'}
- Confirms the agent can recall and apply relevant long-term memory using LlamaIndex in conversations.
- Ensures the MCP ethical guardrails are activated and documented when presented with problematic inputs.
- Verifies that DSPy-optimized prompts lead to measurably better ethical and empathetic responses.
- EmpathyScore target: 0.9
- 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
- Confirms the agent can recall and apply relevant long-term memory using LlamaIndex in conversations.
- Ensures the MCP ethical guardrails are activated and documented when presented with problematic inputs.
- Verifies that DSPy-optimized prompts lead to measurably better ethical and empathetic responses.
Metrics
- EmpathyScore · Target 0.9
- ContextualRelevance · Target 0.85
Workflow tasks
- Task 1
Input: {'user_input': 'string', 'previous_context': ['string']}
Output: {'companion_response': 'string', 'emotional_understanding_score': 'float'}
Execution harness: Python
- Task 2
Input: {'user_input': 'string', 'expected_action': 'string'}
Output: {'ethical_flag_raised': 'boolean', 'intervention_message': 'string'}
Execution harness: Python
Recipe freeze marker
No recipe freeze time is recorded.