Build an LNG Project Knowledge Retrieval System with CrewAI & Milvus
Managing technical specifications for complex LNG projects requires multi-agent document analysis. Build a collaborative multi-agent system using CrewAI and Milvus vector database to index FEED documents, extract environmental specs, and verify engineering compliance.
What you are building
Deploy specialized CrewAI agent roles backed by Milvus vector index to analyze LNG terminal engineering specifications.
How work is evaluated
Evaluates multi-agent CrewAI retrieval and agreement accuracy on LNG technical specs.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
meets_standard_check
Verify compliance result is accurate based on spec standard
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
retrieval_recall
Recall metric of relevant technical document chunks from Milvus • target: 0.96 • range: 0.85-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Orchestrate CrewAI agents with dedicated roles: Engineering Compliance Analyst and Procurement Lead
Implement high-performance vector search using Milvus for technical FEED documentation
Verify LNG cryo-tank insulation and environmental emission compliance standards
Produce consolidated multi-agent audit summary reports
Resources
Engineering specification documents, FEED studies, and regulatory standards for UAE LNG terminals.
Challenge setup
Evaluates multi-agent CrewAI retrieval and agreement accuracy on LNG technical specs.
Challenge input
JSON containing spec query and target standards
CrewAI
Multi-agent coordination for specialized role playback
Milvus
Scalable search over large technical PDF collections
Evaluated output
JSON evaluation report
- Verify compliance result is accurate based on spec standard
- Recall metric of relevant technical document chunks from Milvus • target: 0.96 • range: 0.85-1
- Benchmark: RAGBench-Engineering
- Retrieval Recall target: 0.96
- 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
- CrewAI · Required
- crewAI · Optional
- Milvus · Optional
Evaluation contract
- meets_standard_check · Weight 1
- retrieval_recall · Weight 1
Metrics
- retrieval_recall · Target 0.96
Benchmarks
- RAGBench-Engineering
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing spec query and target standards
Output: JSON evaluation report
Execution harness: Python
Evaluation tools
- CrewAI · Agent Framework
- Milvus · Vector DB
Recipe freeze marker
No recipe freeze time is recorded.