Offshore EPC Bid Evaluation and Risk Agent with Claude Agents SDK
Offshore oilfield expansion tenders involve evaluating multi-billion dollar EPC proposals across complex technical and commercial matrices. Develop an autonomous agent using Claude Agents SDK with extended thinking to parse vendor technical submittals, cross-check compliance against Qatari maritime standards, and calculate risk scores within 5% of benchmark evaluations.
What you are building
The core problem, expected build, and operating context for this challenge.
Build an autonomous technical bid analysis agent using Claude Agents SDK with extended thinking to evaluate offshore oil field EPC proposals.
How work is evaluated
Evaluates technical compliance parsing and commercial risk scoring against benchmark judgments.
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.
non_compliance_flagging
Must flag proposal as non-compliant when jacket depth is insufficient
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
risk_scoring_accuracy
Deviation of generated risk score from human benchmark score • target: 95 • 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
Build an autonomous agent with Claude Agents SDK utilizing Claude 3.7 Sonnet extended thinking
Extract compliance technical specs from complex EPC bid documentation
Cross-examine vendor jacket installation timelines against Qatari offshore safety codes
Generate weighted scorecards for commercial and technical proposal evaluation
Reference links and supporting material
Offshore platform engineering proposals, Qatari regulatory specs, and bid matrix data for Qatar offshore field expansion.
Challenge setup
Evaluates technical compliance parsing and commercial risk scoring against benchmark judgments.
Challenge input
JSON containing vendor proposal details and mandatory technical specs.
Claude Agents SDK
Required Anthropic framework for autonomous extended thinking agents
Evaluated output
JSON containing technical compliance score, identified risk flags, and overall recommendation.
- Must flag proposal as non-compliant when jacket depth is insufficient
- Deviation of generated risk score from human benchmark score • target: 95 • range: 0-100
- Benchmark: Complex Commercial Bid Evaluation Benchmark
- Risk Scoring Accuracy target: 95
- 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
- TPOT · Required
- ElevenLabs · Optional
- Arize Phoenix · Optional
- Arize Phoenix · Optional
Evaluation contract
- non_compliance_flagging · Weight 1
- risk_scoring_accuracy · Weight 1
Metrics
- risk_scoring_accuracy · Target 95
Benchmarks
- Complex Commercial Bid Evaluation Benchmark
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing vendor proposal details and mandatory technical specs.
Output: JSON containing technical compliance score, identified risk flags, and overall recommendation.
Execution harness: Python
Evaluation tools
- Claude Agents SDK · Agent Framework
Recipe freeze marker
No recipe freeze time is recorded.