Integrative Industry Synthesis

Assurance-Centered Agentic AIOps: Parallel Contested Orchestration for Industrial DevSecAIOps

This project implements a new integrated DevSecAIOps synthesis artifact for local-cloud IIoT/OT operations by combining outputs and methods from P3, P4, P5, and P6.

Project repository: https://github.com/Ohara124c41/integrated-industrial-application-acaa

Integrated components

• P3: vulnerability prior context (KEV/NVD signals)
• P4: telemetry risk profile and deep-model-derived prevalence context
• P5: generated RCA narratives with quality/safety tags
• P6: policy-gated multi-agent decision orchestration

Core files

• integrated_system.ipynb: notebook for system execution, EDA/statistics, dimensional analysis, governance checks, and fairness screening
• local_cloud_plugin_architecture.ipynb: thesis-extension notebook for local-cloud plugin orchestration, storage materialization, and failure drills
• integration_runtime.py: integration backend that builds unified packets and evaluation outputs
• contested_orchestration.py: LangChain + LangGraph dual-branch orchestration and meta-adjudication with optional HITL overrides
• local_cloud/: plugin contracts, registry, runtime, and plugin modules for local-cloud control-plane emulation
• data/contracts/*.json: explicit interface contracts
• diagrams/system_architecture.md: architecture flow
• diagrams/threat_model.md: threat model
• outputs/: generated integrated artifacts

Run order

1. Confirm upstream artifacts exist in P3/, P4/, P5/, and P6/.
2. Run notebook: integrated_system.ipynb (recommended).
3. Or run backend directly:

python integration_runtime.py --output-dir outputs
python integration_runtime.py --output-dir outputs --agentic-compare --model gpt-4o-mini

Generated outputs

• outputs/integrated_packets.jsonl
• outputs/system_summary.json
• outputs/failure_cases.json
• outputs/branch_packets.jsonl
• outputs/meta_decisions.jsonl
• outputs/meta_summary.json
• outputs/responsibility_log.jsonl
• outputs/local_cloud_summary.json
• outputs/meta_decisions_local_cloud.jsonl
• outputs/local_cloud_packets.db
• outputs/local_cloud_packets.parquet (optional, if parquet engine available)

Contested orchestration controls

• P7_CONTESTED_LLM=1 enables LLM enrichment for branch rationale text.
• P7_CONTESTED_MODEL=gpt-4o-mini sets the branch-enrichment model name.
• P7_CONTESTED_LLM_PASSES=2 controls multi-pass branch refinement depth (more calls, richer rationale notes).
• P7_CONTESTED_META_LLM=1 enables a meta-level advisory LLM note per incident while keeping deterministic selection gates.
• Meta-selection remains policy-constrained even when branch text is LLM-enriched.
• Hard safety invariant: policy-failing incidents are forced to final_recommendation='escalate'.
• HITL overrides are allowed only within policy bounds; blocked override attempts are logged in meta_decisions*.jsonl.

Notes

• This artifact is intentionally integration-focused and policy-first.
• It reuses prior validated modules and adds a new synthesis/evaluation layer with governance and audit emphasis.
• The local_cloud_plugin_architecture.ipynb notebook is thesis-oriented and can be cited as architecture-extension evidence without changing capstone submission scope.