Power Plant EPC Milestone Monitor with Pydantic AI & Hamming
Contractors confirmed a $683m combined-cycle gas turbine (CCGT) power plant contract in Oman. Build a structured Pydantic AI agent monitored by Hamming to extract contractual milestones from construction logs, identify delay risks, and trigger mitigation tasks. Achieve 90%+ extraction accuracy on vendor progress reports.
What you are building
Develop a type-safe Pydantic AI agent with Hamming evaluation tracking to extract construction milestones and forecast project delay risks.
How work is evaluated
Evaluates milestone extraction accuracy and delay risk detection from construction logs.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
schema_validation
Checks if output conforms strictly to Pydantic model
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
field_extraction_accuracy
Precision of delay days and milestone name extraction • target: 0.9 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Define strict Pydantic schemas for EPC construction milestones and deliverables.
Build a Pydantic AI agent that parses unstructured contractor status reports.
Integrate Hamming for continuous prompt debugging and automated evaluation metrics.
Implement delay detection logic based on critical path dependencies.
Resources
Synthetic contractor status reports modeled after Middle Eastern combined-cycle plant construction logs.
Challenge setup
Evaluates milestone extraction accuracy and delay risk detection from construction logs.
Challenge input
Raw text log from site engineers
Pydantic AI
Provides type-safe structured parsing.
Hamming
Ensures prompt consistency and extraction metrics tracking.
Evaluated output
JSON matching EPCMilestone Pydantic model
- Checks if output conforms strictly to Pydantic model
- Precision of delay days and milestone name extraction • target: 0.9 • range: 0-1
- Benchmark: ExtractBench-EPC
- Field Extraction Accuracy target: 0.9
- 1 public reference case
- 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
- Hamming · Required
- Pydantic AI · Optional
- Lindy · Optional
- Pydantic AI · Optional
Evaluation contract
- schema_validation · Weight 1
- field_extraction_accuracy · Weight 1
Metrics
- field_extraction_accuracy · Target 0.9
Benchmarks
- ExtractBench-EPC
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: Raw text log from site engineers
Output: JSON matching EPCMilestone Pydantic model
Execution harness: Python
Evaluation tools
- Pydantic AI · Agent Framework
- Hamming · Evaluations & Observability
Recipe freeze marker
No recipe freeze time is recorded.