Smelter Restoration Risk Analyzer via Claude Agents SDK & Vercel AI SDK
Emirates Global Aluminium's Al-Taweelah restoration project requires live structural inspection log evaluation to detect bottlenecks. Build an autonomous agent using Claude Agents SDK and Vercel AI SDK to analyze engineering reports, compute completion percentage drift, and generate proactive mitigation steps.
What you are building
The core problem, expected build, and operating context for this challenge.
Combine Claude Agents SDK extended thinking features with Vercel AI SDK UI streaming to construct an aluminium smelter restoration diagnostic system.
How work is evaluated
Validates the accuracy of estimated completion percentage and identified restoration risks from engineering audit texts.
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.
Completion Percentage Test
Validates that completion_percentage correctly equals 25.0.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Risk Classification F1
Accuracy of extracted risk level classification. • target: 1 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Utilize Claude Agents SDK extended reasoning to diagnose smelter restoration risks
Stream real-time inspection log analysis to TypeScript applications using Vercel AI SDK
Parse visual and textual site audit reports from EGA Al-Taweelah restoration milestones
Evaluate progress velocity against target milestone completions
Reference links and supporting material
Synthetic site inspection logs and structural engineering audit notes for Emirates Global Aluminium Al-Taweelah plant restoration.
Challenge setup
Validates the accuracy of estimated completion percentage and identified restoration risks from engineering audit texts.
Challenge input
JSON object with site audit log, completed potlines count, and active structural defect descriptions.
Claude Agents SDK
Extended thinking and visual diagnostic reasoning
Vercel AI SDK
Type-safe tool execution and streaming UI components
Evaluated output
JSON object containing computed completion percentage, identified bottleneck risks, and priority score.
- Validates that completion_percentage correctly equals 25.0.
- Accuracy of extracted risk level classification. • target: 1 • range: 0-1
- Benchmark: GCC Heavy Industrial Audit Benchmark
- Risk Classification F1 target: 1
- 1 public reference case
- JavaScript execution harness
- javascript 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
- Vercel AI SDK · Required
- Ludwig · Optional
- Hamming · Optional
- Vercel AI SDK · Required
Evaluation contract
- Completion Percentage Test · Weight 1
- Risk Classification F1 · Weight 1
Metrics
- Risk Classification F1 · Target 1
Benchmarks
- GCC Heavy Industrial Audit Benchmark
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON object with site audit log, completed potlines count, and active structural defect descriptions.
Output: JSON object containing computed completion percentage, identified bottleneck risks, and priority score.
Execution harness: JavaScript
Evaluation tools
- Claude Agents SDK · Agent Framework
- Vercel AI SDK · UI & Streaming SDK
Recipe freeze marker
No recipe freeze time is recorded.