Global Development Restart Pipeline using OpenAI Agents SDK on Vertex AI
Major global real estate developers like Hines are restarting development pipelines under shifting interest rates. Build an underwriting pipeline using the OpenAI Agents SDK hosted on Google Vertex AI to evaluate project restart triggers across commercial and multifamily properties with 95% metric precision.
What you are building
The core problem, expected build, and operating context for this challenge.
Deploy an OpenAI Agents SDK workflow running on Google Vertex AI runtime that ingests macroeconomic yield spreads and local construction cost indices to recommend development restart timing.
How work is evaluated
Evaluates OpenAI Agents SDK decision logic regarding development project restart triggers.
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.
Restart Recommendation Accuracy
Ensures 'GO' or 'HOLD' recommendation correctly aligns with financial threshold rules.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
yield_on_cost_mae
Mean Absolute Error of calculated Yield on Cost metric • target: 0.001 • range: 0-0.05
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 the OpenAI Agents SDK for creating multi-agent handoffs between market research and underwriting units.
Deploy inference calls via Google Vertex AI model endpoints.
Evaluate Yield-on-Cost (YoC) and capitalization rate spread thresholds.
Generate actionable development restart trigger signals.
Reference links and supporting material
Dataset of 35 real estate development projects detailing estimated ground-up budget, projected NOI, exit capitalization rates, and financing interest rates.
Challenge setup
Evaluates OpenAI Agents SDK decision logic regarding development project restart triggers.
Challenge input
JSON containing projected_stabilized_noi, total_development_cost, exit_cap_rate, and debt_service_rate
OpenAI Agents SDK
Mandatory SDK for defining multi-agent workflows and handoffs.
Google Vertex AI
Required enterprise cloud runtime for hosting model endpoints.
Evaluated output
JSON with yield_on_cost, cap_rate_spread_bps, restart_recommendation ('GO'|'HOLD'), and target_quarter
- Ensures 'GO' or 'HOLD' recommendation correctly aligns with financial threshold rules.
- Mean Absolute Error of calculated Yield on Cost metric • target: 0.001 • range: 0-0.05
- Benchmark: CRE Investment Decision Benchmark
- Yield On Cost Mae target: 0.001
- 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
- Exa · Required
- OpenAI · Optional
- Google Vertex AI · Optional
Evaluation contract
- Restart Recommendation Accuracy · Weight 1
- yield_on_cost_mae · Weight 1
Metrics
- yield_on_cost_mae · Target 0.001
Benchmarks
- CRE Investment Decision Benchmark
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing projected_stabilized_noi, total_development_cost, exit_cap_rate, and debt_service_rate
Output: JSON with yield_on_cost, cap_rate_spread_bps, restart_recommendation ('GO'|'HOLD'), and target_quarter
Execution harness: Python
Evaluation tools
- OpenAI Agents SDK · Agent Framework
- Google Vertex AI · Inference Infrastructure
Recipe freeze marker
No recipe freeze time is recorded.