What the presenter says
"Notice that every AI-facing step is constrained. HARP retrieves evidence and safe diagnostics. Nero tracks the decision. LiteLLM can summarize, but it does not replace citations. The architecture keeps the human in control."
Back to HARP
Interactive Simulation
Alert to evidence to human decision.
This walkthrough shows how the stack behaves when a Kubernetes alert appears. Every AI-facing step is constrained: HARP retrieves evidence, Nero tracks decisions, and humans approve risky action.
What the audience sees
HARP result: docs/cluster-operations/runbook-wordpress-pvc-storage-fix.md
Nero contract: docs/contracts/nero-evidence-contract.md
RAG contract: docs/specs/rag-connector-contract.md
Why This Works
The loop creates reusable operational memory.
01
Observe
Monitoring and humans produce a signal.
02
Retrieve
HARP finds prior context from runbooks and RCAs.
03
Ground
Evidence pack contains citations, snippets, confidence, and gaps.
04
Approve
Nero-Camp records human decisions for risky action.
05
Document
Outcome becomes an RCA, runbook update, or session log.
06
Index
OpenRAG and HARP learn from the new document.