Skip to main content
Back to challenges
Challenge

Adaptive Reasoning with Gemini 3 Flash: A LangGraph MCP Agent for Cost-Optimized Analytics

This challenge focuses on building an intelligent agent system. You will design and implement a graph-based workflow using LangGraph that dynamically switches between Gemini 3 Flash (for speed and cost-efficiency) and Gemini 2.5 Pro (for deeper, more complex reasoning tasks). The system must employ MCP-enabled tool integration to access enterprise data sources and make informed decisions on model selection, ensuring optimal resource allocation for analytical tasks like market trend analysis or customer feedback summarization. The goal is to achieve high-quality insights while meticulously managing operational costs.

Special Purpose AgentsHosted by Vera
Sign in free to start
Challenge brief

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 DSPy, leveraging Gemini 3 Flash for adaptive, cost-optimized reasoning through MCP tool integration.

Delivery guide

How work is evaluated

Evaluation

The evaluation will assess the agent's ability to effectively leverage adaptive reasoning, demonstrating judicious model selection, prompt optimization, and accurate tool integration, all while balancing cost and performance metrics.

Datasets

Challenge datasets

Public datasets, plus the private datasets you have access to.

Learning goals

What you should walk away with

  • Design and implement adaptive reasoning logic for dynamic LLM selection.

  • Build complex graph-based agent workflows using LangGraph.

  • Evaluate and compare cost, latency, and reasoning quality across different Gemini models.

  • Implement MCP-enabled tool integration for enterprise data access.

  • Optimize agent prompts and reasoning steps with DSPy.

Resources and assets

Reference links and supporting material

Dataset notes

Sample data for 2 tasks

Challenge setup

The evaluation will assess the agent's ability to effectively leverage adaptive reasoning, demonstrating judicious model selection, prompt optimization, and accurate tool integration, all while balancing cost and perf...

Challenge input

{'market_reports': ['report1_text', 'report2_text'], 'news_articles': ['article1_text', 'article2_text']}

Agent execution

No agent tool is listed for this challenge.

Evaluated output

{'summary': 'string', 'estimated_model_cost': 'float', 'latency_ms': 'int', 'models_used': ['Gemini 3 Flash', 'Gemini 2.5 Pro']}

Evaluation checks
  • Agent must correctly identify and utilize the appropriate Gemini model (Flash vs. Pro) based on task comple...
  • Agent must successfully invoke MCP-enabled tools for data retrieval and contextual reasoning, demonstrating...
  • The agent's prompts should demonstrate evidence of DSPy optimization (e.g., programmatic decomposition, few...
Configured evaluation evidence
  • Cost Efficiency (USD Per Task) target: 0.008
Configured runtime
  • 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

  • Agent must correctly identify and utilize the appropriate Gemini model (Flash vs. Pro) based on task complexity and explicit instructions (e.g., 'deep dive' tasks should trigger Pro).
  • Agent must successfully invoke MCP-enabled tools for data retrieval and contextual reasoning, demonstrating correct parameter passing and response handling.
  • The agent's prompts should demonstrate evidence of DSPy optimization (e.g., programmatic decomposition, few-shot examples, chain-of-thought where applicable), leading to higher quality outputs.

Metrics

  • Cost-Efficiency (USD per task) · Target 0.008
  • Latency (ms per task) · Target 2500
  • Reasoning Accuracy · Target 4.5

Workflow tasks

  1. Task 1

    Input: {'market_reports': ['report1_text', 'report2_text'], 'news_articles': ['article1_text', 'article2_text']}

    Output: {'summary': 'string', 'estimated_model_cost': 'float', 'latency_ms': 'int', 'models_used': ['Gemini 3 Flash', 'Gemini 2.5 Pro']}

    Execution harness: Python

  2. Task 2

    Input: {'financial_statement': 'text', 'economic_indicators': {'gdp_growth': 'float', 'inflation_rate': 'float'}}

    Output: {'risk_assessment': 'string', 'mitigation_strategies': ['strategy1', 'strategy2'], 'estimated_model_cost': 'float', 'latency_ms': 'int', 'models_used': ['Gemini 3 Flash', 'Gemini 2.5 Pro']}

    Execution harness: Python

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Adaptive Reasoning with Gemini 3 Flash: A LangGraph MCP Agent for Cost-Optimized Analytics