Gemini 2.5 Pro Driving Co-Pilot with LangGraph & Hybrid Reasoning
This challenge focuses on building a sophisticated, context-aware driving co-pilot. Developers will design a multi-modal agent system that leverages Gemini 2.5 Pro's advanced conversational capabilities and LangGraph's robust state management to provide real-time, adaptive assistance. The co-pilot will handle complex queries, offer proactive suggestions based on live data, and maintain situational awareness, moving beyond simple turn-by-turn navigation to a truly intelligent driving companion. The core of the system will involve implementing a hybrid instant/deep reasoning architecture. For routine inquiries like 'Where's the nearest gas station?', the agent will use instant reasoning. For complex, multi-step planning or critical safety assessments, it will engage Gemini 2.5 Pro's Deep Think mode, dynamically allocating higher computational resources. Tool integration via a lightweight Model Context Protocol (MCP) will enable seamless access to real-time traffic, weather, and location-based services, making the agent highly practical for real-world applications.
What you are building
The core problem, expected build, and operating context for this challenge.
Build a graph-based multi-modal agent using LangGraph and Gemini 2.5 Pro for conversational, context-aware driving assistance with hybrid reasoning.
How work is evaluated
The evaluation will assess the co-pilot's ability to provide accurate, timely, and context-aware assistance, its conversational fluency, and its effective use of hybrid reasoning and tool integration in simulated driving scenarios.
Challenge datasets
Public datasets, plus the private datasets you have access to.
What you should walk away with
Develop strategies for managing adaptive thinking budgets in dynamic environments.
Integrate real-time external tools and APIs for location and traffic data.
Master LangGraph for building stateful conversational agent workflows.
Implement multi-modal input and output processing with Gemini 2.5 Pro.
Design and deploy hybrid instant/deep reasoning for adaptive decision-making.
Reference links and supporting material
Sample data for 2 tasks
Challenge setup
The evaluation will assess the co-pilot's ability to provide accurate, timely, and context-aware assistance, its conversational fluency, and its effective use of hybrid reasoning and tool integration in simulated driv...
Challenge input
{ "user_query": "string", "current_location": "lat,lon", "destination": "string", "traffic_conditions": "enum<light|moderate|heavy>" }
Agent execution
No agent tool is listed for this challenge.
Evaluated output
{ "navigation_plan": "string", "estimated_time": "string", "detour_suggestion": "string", "reasoning_mode_used": "enum<instant|deep>" }
- Navigation plan must be logically sound and address traffic conditions.
- Complex tasks should trigger 'deep' reasoning, simple tasks 'instant'.
- Conversational Fluency 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
- Navigation plan must be logically sound and address traffic conditions.
- Complex tasks should trigger 'deep' reasoning, simple tasks 'instant'.
Metrics
- Conversational Fluency Score · Target 4
- Tool Integration Latency (ms) · Target 300
Workflow tasks
- Task 1
Input: { "user_query": "string", "current_location": "lat,lon", "destination": "string", "traffic_conditions": "enum<light|moderate|heavy>" }
Output: { "navigation_plan": "string", "estimated_time": "string", "detour_suggestion": "string", "reasoning_mode_used": "enum<instant|deep>" }
Execution harness: Python
- Task 2
Input: { "user_query": "string", "current_location": "lat,lon", "visual_context_present": "boolean", "poi_type": "string" }
Output: { "poi_name": "string", "address": "string", "distance": "string", "map_display_action": "string", "reasoning_mode_used": "enum<instant|deep>" }
Execution harness: Python
Recipe freeze marker
No recipe freeze time is recorded.