Recommender Multi-Agent for Complex Suggestions
This challenge tasks you with building a cutting-edge multi-agent recommendation system using CrewAI. Your system will leverage GPT-5 for highly creative and nuanced recommendation generation, supported by Claude Sonnet 4 for robust information retrieval and synthesis. Agents will collaborate to understand complex user intent, retrieve obscure related data via RAG, and generate highly personalized, multi-faceted recommendations for music or other domains. A critical component is the integration of MCP-enabled tools to access external data sources (e.g., music catalogs, user preference databases, external reviews) to enrich the recommendation process and provide context beyond the LLM's initial training data. The system must demonstrate advanced extended thinking capabilities, adapting its reasoning budget based on recommendation complexity.
What you are building
The core problem, expected build, and operating context for this challenge.
Build a CrewAI-orchestrated multi-agent system leveraging GPT-5 and Claude Sonnet 4 for complex, personalized recommendations, utilizing RAG and MCP-enabled tools.
How work is evaluated
The evaluation will assess the recommendation system's ability to handle complex queries, retrieve obscure information, generate creative and relevant recommendations, and demonstrate effective agent collaboration and tool usage. Focus on relevance, creativity, factual accuracy, and the depth of reasoning.
Challenge datasets
Public datasets, plus the private datasets you have access to.
What you should walk away with
Develop RAG pipelines for retrieving obscure domain-specific information.
Apply extended thinking with adaptive reasoning budgets for nuanced outputs.
Implement MCP-enabled tool integration for external data access.
Integrate GPT-5 and Claude Sonnet 4 for diverse reasoning and content generation.
Design and implement a CrewAI multi-agent system for complex recommendations.
Reference links and supporting material
Sample data for 1 tasks
Challenge setup
The evaluation will assess the recommendation system's ability to handle complex queries, retrieve obscure information, generate creative and relevant recommendations, and demonstrate effective agent collaboration and...
Challenge input
{ "user_profile": { "age": 30, "mood": "reflective", "fav_genres": ["indie folk", "ambient electronic"], "recent_activities": ["reading philosophy"...
Agent execution
No agent tool is listed for this challenge.
Evaluated output
{ "recommendation_title": "string", "artist": "string", "genre_tags": ["string"], "reasoning": "string (detailed explanation of why this recommenda...
- Verifies that the output is valid JSON and matches the specified schema.
- Confirms that the MCP-enabled tools were invoked successfully and their outputs were utilized in the recomm...
- Recommendation Relevance Score target: 4
- 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
- Verifies that the output is valid JSON and matches the specified schema.
- Confirms that the MCP-enabled tools were invoked successfully and their outputs were utilized in the recommendation process (e.g., 'external_sources_used' populated).
Metrics
- Recommendation Relevance Score · Target 4
- Creativity and Nuance Score · Target 4
- Reasoning Clarity and Detail · Target 4
Workflow tasks
- Task 1
Input: { "user_profile": { "age": 30, "mood": "reflective", "fav_genres": ["indie folk", "ambient electronic"], "recent_activities": ["reading philosophy", "stargazing"] }, "query": "I'm looking for a new album that combines introspective lyrics with atmospheric soundscapes, something that pairs well with late-night contemplation and has a subtle nod to early 2000s indie electronica, perhaps from an artist not widely known." }
Output: { "recommendation_title": "string", "artist": "string", "genre_tags": ["string"], "reasoning": "string (detailed explanation of why this recommendation fits the query, referencing obscure elements and user profile)", "external_sources_used": [{"tool_name": "string", "data_retrieved": "string"}] }
Execution harness: Python
Recipe freeze marker
No recipe freeze time is recorded.