FinTech Credit Advisor
The intersection of FinTech and AI offers immense opportunities for personalized financial guidance. This challenge involves building an intelligent credit building advisor using a graph-based multi-agent system. The advisor will leverage Gemini 3 Pro's multimodal capabilities and hybrid reasoning to analyze a user's simulated financial data. The goal is to provide actionable, personalized advice for improving credit scores. The system will be orchestrated using LangGraph to define dynamic, stateful financial advisory workflows. Agents will communicate using an A2A protocol (simulated) for seamless data exchange and decision-making. Key features include MCP tool integration to connect with simulated financial data APIs (bank statements), adaptive reasoning budgets for complex financial scenarios, and the generation of tailored credit improvement plans.
What you are building
The core problem, expected build, and operating context for this challenge.
Develop a graph-based multi-agent FinTech advisor using LangGraph, Gemini 3 Pro, and MCP for personalized credit building advice, integrated with simulated financial APIs.
How work is evaluated
The evaluation will assess the FinTech advisor system's ability to accurately analyze simulated financial data, provide personalized and relevant credit-building advice, and correctly utilize MCP tool integrations. Emphasis will be on the coherence of the LangGraph workflow and the actionable quality of the recommendations.
Challenge datasets
Public datasets, plus the private datasets you have access to.
What you should walk away with
Build complex stateful agent workflows using LangGraph.
Utilize Gemini 2.5 Pro for hybrid reasoning on financial data.
Integrate multiple external financial APIs via MCP.
Generate personalized, actionable financial advice.
Implement A2A protocol for internal agent communication.
Reference links and supporting material
Sample data for 2 tasks
Challenge setup
The evaluation will assess the FinTech advisor system's ability to accurately analyze simulated financial data, provide personalized and relevant credit-building advice, and correctly utilize MCP tool integrations. Em...
Challenge input
{'user_profile': {'income': float, 'expenses': float, 'debt': float, 'credit_score': int}, 'rent_history': [{'month': 'YYYY-MM', 'paid_on_time': bo...
Agent execution
No agent tool is listed for this challenge.
Evaluated output
{'credit_plan_summary': 'string', 'action_items': [{'step': int, 'description': 'string', 'expected_impact': 'string'}], 'mcp_tool_calls': ['list o...
- Output is valid JSON and matches schema.
- At least one MCP tool call is detected and relevant to the task.
- Recommendations are specific, actionable steps (manual spot check).
- RelevanceOfAdvice 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
- Output is valid JSON and matches schema.
- At least one MCP tool call is detected and relevant to the task.
- Recommendations are specific, actionable steps (manual spot check).
- Advice is tailored to the simulated user profile/data.
Metrics
- RelevanceOfAdvice · Target 0.9
- WorkflowCoherence · Target 0.85
- DataIntegrationAccuracy · Target 0.9
Workflow tasks
- Task 1
Input: {'user_profile': {'income': float, 'expenses': float, 'debt': float, 'credit_score': int}, 'rent_history': [{'month': 'YYYY-MM', 'paid_on_time': bool}]}
Output: {'credit_plan_summary': 'string', 'action_items': [{'step': int, 'description': 'string', 'expected_impact': 'string'}], 'mcp_tool_calls': ['list of tool calls']}
Execution harness: Python
- Task 2
Input: {'transactions': [{'date': 'YYYY-MM-DD', 'amount': float, 'category': 'string'}], 'rent_history': [{'month': 'YYYY-MM', 'paid_on_time': bool}]}
Output: {'analysis_report': 'string', 'insights': ['string'], 'mcp_tool_calls': ['list of tool calls']}
Execution harness: Python
Recipe freeze marker
No recipe freeze time is recorded.