Google Cloud Rapid Agent Hackathon  ·  GitLab + Arize Phoenix tracks

AI SRE that fixes
Kubernetes SLO breaches
in 60 seconds

NeuroScale Agents detects a P99 latency breach via Arize Phoenix MCP, diagnoses root cause using RAG over SRE runbooks, and opens a Kyverno-compliant GitLab Merge Request — autonomously.

▶  Watch 3-min Demo ★  GitHub Repo 🖥  Live Dashboard
< 60s
Anomaly → GitLab MR
7/7
Verification tests pass
3
Agents (A2A pattern)
2
Partner MCP integrations
$0
Credentials to run demo

Live Demo

Anomaly detected → root cause diagnosed → GitLab MR opened. Problem, architecture, pipeline, live terminal. Real agent output.

NeuroScale Agents — 3-Minute Demo thumbnail

Three Agents. Two MCPs. One Closed Loop.

The only submission using both Arize Phoenix and GitLab MCP in a single autonomous pipeline.

Arize Phoenix MCP → Watcher Agent → Diagnostician Agent → Operator Agent → GitLab MR → ArgoCD Sync
👁
Watcher Agent
Polls Arize Phoenix via MCP. Reads span metrics, scores severity — CRITICAL / WARNING / INFO. Fires structured incident report to Diagnostician.
get_spans get_trace
🔬
Diagnostician Agent
TF-IDF search over SRE runbooks. Classifies root cause. Generates Kyverno-compliant YAML patch validated against policy constraints.
CPU_THROTTLING MODEL_DRIFT RESOURCE_EXHAUSTION rag_store.py
⚙️
Operator Agent
Opens GitLab MR with full root cause analysis, confidence score, Kyverno checklist. Then stops — waits for human approval.
create_branch create_or_update_file create_merge_request list_merge_requests

Before vs. After

Metric Manual SRE NeuroScale Agents
Detection → MR open 30–90 minutes < 60 seconds
On-call pages Every incident Approval only
Runbook compliance Ad-hoc, memory-dependent 100% RAG-enforced
Kyverno policy coverage Manual audit, post-hoc Every MR, automatically
Cost per incident $100K+/hr Agent compute only
MTTR 30–90 min 5–10 min (post-approval)

Governance Hardwired In

The AI doesn't bypass your policies. Every generated YAML patch is built to satisfy Kyverno policy constraints. The MR description includes a full compliance checklist — human reviewer sees exactly what was checked. Fails safe, not silently.

✓ ENFORCED
Resource limits — require-resource-requests-limits
✓ ENFORCED
No root containers — disallow-root-containers
✓ ENFORCED
Cost center labels — require-standard-labels
✓ ENFORCED
Rolling update strategy mandated

Try It — Zero Credentials Required

Runs locally. No GCP account. No GitLab token. No Arize API key. Full pipeline in under 60 seconds.

bash — neuroscale-agents-v2
# Clone and run — no credentials needed
git clone https://github.com/sodiq-code/neuroscale-agents-v2
cd neuroscale-agents-v2
pip install httpx scikit-learn

# Verify all 7 agents pass
bash scripts/verify-all.sh
✓ Watcher Agent ......... PASS
✓ Diagnostician Agent ... PASS
✓ Operator Agent ........ PASS
✓ A2A Protocol .......... PASS
✓ GitLab MCP ............ PASS
✓ Arize Phoenix MCP ..... PASS
✓ Kyverno Validation .... PASS
→ 7/7 PASS

# Run the full pipeline: anomaly → GitLab MR
bash scripts/demo-run.sh
🔍 Watcher: CRITICAL P99 latency SLO breach detected
🔬 Diagnostician: Root cause = CPU_THROTTLING (confidence: 0.94)
⚙️ Operator: GitLab MR #47 opened — awaiting human approval
✓ Pipeline complete in 58 seconds

Built With

A2A Pattern
Multi-agent orchestration
Gemini 1.5 Pro
Root cause reasoning
Arize Phoenix
ML observability MCP
GitLab MCP
GitOps automation
TF-IDF RAG
Runbook retrieval
Kyverno
Policy governance
KServe
Inference serving
ArgoCD
GitOps deployment
GKE
Kubernetes runtime
A2A Protocol
Agent-to-agent comms