Skip to main content
Loading navigation
Back to challenges

Multi-Modal Vessel Loading Inspection Agent with Claude SDK and Gemini 2.5 Flash

Maritime crude oil loading requires dynamic inspection of vessel draft marks, loading arm alignment, and safety compliance images at marine terminals. Build a multimodal vision-language agent using Claude Agents SDK and Gemini 2.5 Flash as high-speed secondary vision runtime. The agent analyzes port surveillance photos, checks safety compliance against international tanker standards (ISGOTT), and calculates estimated laytime remaining based on visual pump connections.

Business OperationsHosted by Vera
Sign in free to start

What you are building

Develop a multimodal marine terminal inspection agent using Claude Agents SDK and Gemini 2.5 Flash.

How work is evaluated

Evaluation

Evaluates visual terminal inspection agent on safety flag identification.

Datasets

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

How submissions are scored

Dimensions
2 scoring checks
Binary
2 pass or fail dimensions
Ordinal
0 scaled dimensions
Dimension 1

safety_clearance_passed

Verify clearance is granted when sensors are normal.

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Dimension 2

issue_detection_recall

Recall metric for detecting visual safety violations. • target: 0.95 • range: 0-1

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Learning goals

  • Implement multimodal tool execution with Claude Agents SDK.

  • Incorporate Gemini 2.5 Flash for high-speed image preprocessing and feature extraction.

  • Automate laytime and safety compliance auditing from visual feed streams.

Resources

Dataset notes

Synthetic image metadata and terminal sensor frames from marine crude loading berths.

Challenge setup

Evaluates visual terminal inspection agent on safety flag identification.

Challenge input

JSON containing image_metadata and sensor_readings

Claude Agents SDK

Provides deep reasoning and tool invocation framework.

Gemini 2.5 Flash

Fast visual feature extraction and image processing runtime.

Evaluated output

JSON with safety_clearance, detected_issues, estimated_flow_rate_bph

Evaluation checks
  • Verify clearance is granted when sensors are normal.
  • Recall metric for detecting visual safety violations. • target: 0.95 • range: 0-1
Configured evaluation evidence
  • Benchmark: Vision-Logistics-Bench
  • Issue Detection Recall target: 0.95
  • 1 public reference case
Configured runtime
  • Python execution harness
  • Python sandbox (unavailable on Versalist)

Shows the first task, up to three tools, and up to three evidence items.

View technical recipe

Configured tools

Environment
  • Modal · Optional
Action Space
  • Google Gemini · Required
Policy Serving
  • Gemini 2.5 Flash · Optional
  • Modal · Optional

Evaluation contract

  • safety_clearance_passed · Weight 1
  • issue_detection_recall · Weight 1

Metrics

  • issue_detection_recall · Target 0.95

Benchmarks

  • Vision-Logistics-Bench
  • general

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON containing image_metadata and sensor_readings

    Output: JSON with safety_clearance, detected_issues, estimated_flow_rate_bph

    Execution harness: Python

Evaluation tools

  • Claude Agents SDK · Reasoning Core
  • Gemini 2.5 Flash · Vision Model

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Multi-Modal Vessel Loading Inspection Agent with Claude SDK and Gemini 2.5 Flash