Skip to main content
Loading navigation
Back to challenges

Build a Gas Plant Expansion Risk Analysis Agent with Claude SDK and DeepSeek R1

Megaproject gas developments present layered supply chain and technical execution risks. Build a reasoning agent using Claude Agents SDK and DeepSeek R1 to analyze EPC procurement schedules for Jafurah Phase 4, flagging critical bottleneck dependencies with over 90% risk detection coverage.

Business OperationsHosted by Vera
Sign in free to start

What you are building

Orchestrate reasoning and tool calling with Claude Agents SDK and DeepSeek R1 for deep EPC risk extraction.

How work is evaluated

Evaluation

Evaluates EPC risk identification accuracy across Jafurah expansion phase milestones.

Datasets

Public datasets, plus the private datasets you have access to.

How submissions are scored

Dimensions
2 scoring checks
Binary
2 pass or fail dimensions
Ordinal
0 scaled dimensions
Dimension 1

bottleneck_detection_test

Ensures primary bottleneck comp_01 is captured

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Dimension 2

risk_recall

Percentage of true risk bottlenecks correctly identified • target: 0.9 • range: 0-1

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Learning goals

  • Build reasoning chains for industrial risk using Claude Agents SDK

  • Query DeepSeek R1 reasoning endpoints for deep mathematical risk path analysis

  • Parse multi-contract EPC delivery schedules for Jafurah gas expansion project

  • Generate actionable risk mitigation matrices for high-lead long-item procurement

Resources

Dataset notes

Synthetic EPC contract and schedule data modeled after Aramco Jafurah Unconventional Gas Field development phases.

Challenge setup

Evaluates EPC risk identification accuracy across Jafurah expansion phase milestones.

Challenge input

JSON array of schedule_items with lead_time_weeks, dependency_ids, vendor_country

Claude Agents SDK

Extended thinking and tool integration

DeepSeek R1

Deep mathematical path reasoning

Evaluated output

JSON containing high_risk_items, max_project_delay_weeks, mitigation_recommendations

Evaluation checks
  • Ensures primary bottleneck comp_01 is captured
  • Percentage of true risk bottlenecks correctly identified • target: 0.9 • range: 0-1
Configured evaluation evidence
  • Benchmark: Complex Reasoning Benchmark
  • Risk Recall target: 0.9
  • 1 public reference case
Configured runtime
  • 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

Action Space
  • DeepSeek · Required
Observation
  • Gentrace · Optional
Reward / Eval
  • Gentrace · Optional
Policy Serving
  • DeepSeek R1 · Optional

Evaluation contract

  • bottleneck_detection_test · Weight 1
  • risk_recall · Weight 1

Metrics

  • risk_recall · Target 0.9

Benchmarks

  • Complex Reasoning Benchmark
  • general

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON array of schedule_items with lead_time_weeks, dependency_ids, vendor_country

    Output: JSON containing high_risk_items, max_project_delay_weeks, mitigation_recommendations

    Execution harness: Python

Evaluation tools

  • Claude Agents SDK · Agent Orchestrator
  • DeepSeek R1 · Reasoning Model Engine

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Build a Gas Plant Expansion Risk Analysis Agent with Claude SDK and DeepSeek R1