Policy-Aware Content Curator
Build an advanced agentic system that can intelligently curate content from the web, similar to how AI browsers handle paywalls or content restrictions. The system must process news articles, summarize them, and apply dynamic content policies (e.g., avoiding sources from a blacklist, prioritizing specific types of reporting, or finding alternative summaries if a paywall is detected). This involves creating a graph-based workflow where different nodes (agents) handle content fetching, policy checking, summarization, and alternative sourcing. The core of this challenge is to leverage Gemini 2.5 Pro's Deep Think mode for complex policy interpretation and nuanced summarization, particularly when adapting to a 'blacklist' or 'preference list' of sources. Implement an adaptive reasoning budget within the LangGraph workflow, allowing the system to spend more computational effort on articles that trigger complex policy decisions or require extensive alternative source discovery.
What you are building
The core problem, expected build, and operating context for this challenge.
Build a graph-based multi-agent system using LangGraph and Gemini 2.5 Pro Deep Think mode for autonomous, policy-aware content curation, incorporating adaptive reasoning budgets.
How work is evaluated
The evaluation will assess the system's ability to accurately curate content, adhere to dynamic policies, utilize adaptive reasoning, and generate high-quality summaries and alternative reports.
Challenge datasets
Public datasets, plus the private datasets you have access to.
What you should walk away with
Implement advanced RAG techniques for policy document retrieval and contextual content generation.
Master LangGraph for building stateful DAG agent workflows with dynamic control flow.
Build robust tool integration for web scraping, search APIs, and content summarization.
Design and apply adaptive reasoning budgets with Gemini 2.5 Pro's Deep Think mode for nuanced policy interpretation.
Reference links and supporting material
Sample data for 2 tasks
Challenge setup
The evaluation will assess the system's ability to accurately curate content, adhere to dynamic policies, utilize adaptive reasoning, and generate high-quality summaries and alternative reports.
Challenge input
{"article_url": "string", "policy_rules": "list of strings", "blacklist_domains": "list of strings"}
Agent execution
No agent tool is listed for this challenge.
Evaluated output
{"action_taken": "string (summarized|alternative_found|flagged)", "summary": "string", "alternative_source_url": "string (optional)", "policy_viola...
- Ensure no blacklisted sources are directly summarized without alternative action.
- Verify summaries are concise, accurate, and relevant to the original article.
- Confirm that complex scenarios trigger higher resource usage (more API calls/longer processing) than simple...
- Summary Factual Accuracy (F1 Score) 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
- Ensure no blacklisted sources are directly summarized without alternative action.
- Verify summaries are concise, accurate, and relevant to the original article.
- Confirm that complex scenarios trigger higher resource usage (more API calls/longer processing) than simple ones, as evidenced by 'AdaptiveBudgetEfficiencyTest'.
Metrics
- Summary Factual Accuracy (F1-score) · Target 0.9
- Policy Rule Compliance Rate · Target 0.95
- Resource Utilization Ratio (Complex vs. Simple) · Target 3
Workflow tasks
- Task 1
Input: {"article_url": "string", "policy_rules": "list of strings", "blacklist_domains": "list of strings"}
Output: {"action_taken": "string (summarized|alternative_found|flagged)", "summary": "string", "alternative_source_url": "string (optional)", "policy_violation_reason": "string (optional)"}
Execution harness: Python
- Task 2
Input: {"article_url": "string", "policy_rules": "list of strings", "complexity_level": "string (low|medium|high)"}
Output: {"api_calls_made": "integer", "processing_time_ms": "integer", "reasoning_path": "list of strings (LangGraph node sequence)"}
Execution harness: Python
Recipe freeze marker
No recipe freeze time is recorded.