Assess Office Structural Reality using Claude Agents SDK and Fixie
Converting aging office buildings to residential space fails when engineering teams miss structural constraints like deep floorplates and plumbing risers. Build a Claude Agents SDK system with Fixie integration to analyze architectural metrics and output conversion viability scores with 90% accuracy.
What you are building
The core problem, expected build, and operating context for this challenge.
Develop an automated structural evaluation agent using Claude Agents SDK and Fixie to analyze office building floorplates and MEP infrastructure for conversion potential.
How work is evaluated
Evaluates office-to-residential viability score calculation accuracy based on architectural parameters.
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.
Blocker Identification Test
Validates that all architectural constraints exceeding thresholds are identified.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Viability Score Accuracy
Absolute error margin from ground truth score. • target: 5 • 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
Implement Claude Agents SDK tool definitions for parsing floorplate depth, window distance, and riser spacing.
Connect Fixie assistant hooks to enable voice/chat query interface for structural engineers.
Calculate a conversion viability index based on architectural parameters.
Reference links and supporting material
Dataset of 50 office building architectural schemas with floorplate depths, ceiling heights, and conversion outcome labels.
Challenge setup
Evaluates office-to-residential viability score calculation accuracy based on architectural parameters.
Challenge input
JSON with core_depth_ft, floor_to_floor_height_ft, window_perimeter_ratio
Claude Agents SDK
Assigned agent framework for reasoning and tool calling.
Fixie
Provides interactive conversational endpoint and agent integration.
Evaluated output
JSON with viability_score (0-100) and critical_blockers list
- Validates that all architectural constraints exceeding thresholds are identified.
- Absolute error margin from ground truth score. • target: 5 • range: 0-100
- Benchmark: Structural_Conversion_Eval
- Viability Score Accuracy target: 5
- 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
- Fixie · Required
- RAI · Optional
- Braintrust · Optional
- Braintrust · Optional
Evaluation contract
- Blocker Identification Test · Weight 1
- Viability Score Accuracy · Weight 1
Metrics
- Viability Score Accuracy · Target 5
Benchmarks
- Structural_Conversion_Eval
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON with core_depth_ft, floor_to_floor_height_ft, window_perimeter_ratio
Output: JSON with viability_score (0-100) and critical_blockers list
Execution harness: Python
Evaluation tools
- Claude Agents SDK · Agent Framework
- Fixie · Assistant Platform
Recipe freeze marker
No recipe freeze time is recorded.