Emergency Refinery Damage Assessment Agent with Google ADK and Gemini
Refineries facing emergency physical damage require fast damage mapping from visual drone feeds and structural engineering logs. Build a multimodal diagnostic agent using Google ADK and Gemini to parse drone images and repair logs, generating prioritized bill-of-materials and schedule repair Gantt charts with 92% diagnostic coverage.
What you are building
The core problem, expected build, and operating context for this challenge.
Develop a multimodal emergency repair agent using Google ADK and Gemini to analyze visual structural damage and issue repair schedules for oil refineries.
How work is evaluated
Evaluates visual damage detection accuracy and repair plan completeness for refinery assets.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
critical_damage_identification
Must detect secondary cooling line defect in output
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
diagnostic_coverage
Percentage of actual site damages correctly extracted by agent • target: 92 • range: 0-100
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Build a multimodal agent with Google ADK (Agent Development Kit) utilizing Gemini 1.5 Pro
Process thermal and visual drone imagery of refinery distillation towers to map structural damage
Synthesize maintenance repair logs into prioritized Samsung E&A repair Gantt schedules
Validate safety compliance requirements prior to authorizing unit restart steps
Reference links and supporting material
Synthetic drone damage photos and structural engineering logs based on Mina Al-Ahmadi refinery repair reports.
Challenge setup
Evaluates visual damage detection accuracy and repair plan completeness for refinery assets.
Challenge input
JSON containing base64 drone inspection image and raw engineering report text.
Google ADK
Required Google agent development framework
Gemini 1.5 Pro
Required multimodal vision-language model
Evaluated output
JSON containing identified damaged components, estimated repair hours, and bill of materials.
- Must detect secondary cooling line defect in output
- Percentage of actual site damages correctly extracted by agent • target: 92 • range: 0-100
- Benchmark: Multimodal Industrial Inspection Benchmark
- Diagnostic Coverage target: 92
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
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
- Google ADK · Required
- Google Gemini · Optional
- Google · Optional
Evaluation contract
- critical_damage_identification · Weight 1
- diagnostic_coverage · Weight 1
Metrics
- diagnostic_coverage · Target 92
Benchmarks
- Multimodal Industrial Inspection Benchmark
- robotics
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing base64 drone inspection image and raw engineering report text.
Output: JSON containing identified damaged components, estimated repair hours, and bill of materials.
Execution harness: Python
Evaluation tools
- Google ADK · Agent Framework
- Gemini 1.5 Pro · Model Runtime
Recipe freeze marker
No recipe freeze time is recorded.