Purdue Model OT/IT Industrial Knowledge Graph RAG with LlamaIndex
Plant security and control data structured across Level 0 (sensors) to Level 4 (ERP) under the Purdue Model is difficult to query safely without crossing network boundary segments. Build a context-aware RAG pipeline using LlamaIndex to query cross-layer OT/IT architecture documents with 100% compliance with security level access rules.
What you are building
Implement a Purdue-model aware RAG system using LlamaIndex to allow safe cross-layer queries across OT operational zones and IT enterprise levels.
How work is evaluated
Evaluates security boundary filtering compliance and query answer precision across Purdue model layers.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
Boundary Block Verification
Confirms that Level 4 user attempting Level 1 direct access is blocked.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
authorized_query_precision
Accuracy score on valid, allowed cross-level queries. • target: 0.95 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Construct a hierarchical Property Graph Index in LlamaIndex representing Purdue Enterprise levels (0 through 4)
Implement role-based metadata filtering to prevent unauthorized data movement across Level 2 OT / Level 4 IT boundaries
Build dynamic SQL/Vector hybrid retrieval tools targeting plant historian and ERP schemas
Resources
Documents describing manufacturing execution systems (MES), SCADA topologies, and enterprise network architectures classified by Purdue model layers.
Challenge setup
Evaluates security boundary filtering compliance and query answer precision across Purdue model layers.
Challenge input
JSON containing user_role, user_purdue_level, and query_text
LlamaIndex
Offers property graph indices and customizable node postprocessor pipelines.
Evaluated output
JSON with response, boundary_violation_blocked, and retrieved_nodes_count
- Confirms that Level 4 user attempting Level 1 direct access is blocked.
- Accuracy score on valid, allowed cross-level queries. • target: 0.95 • range: 0-1
- Benchmark: OT Security Context RAG Benchmark
- Authorized Query Precision target: 0.95
- 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
- Llama Index · Required
- LlamaIndex · Optional
- Llama 3.2 · Optional
Evaluation contract
- Boundary Block Verification · Weight 1
- authorized_query_precision · Weight 1
Metrics
- authorized_query_precision · Target 0.95
Benchmarks
- OT Security Context RAG Benchmark
- security
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing user_role, user_purdue_level, and query_text
Output: JSON with response, boundary_violation_blocked, and retrieved_nodes_count
Execution harness: Python
Evaluation tools
- LlamaIndex · Data Framework & RAG Engine
Recipe freeze marker
No recipe freeze time is recorded.