Network Operations Copilot
Outcome
Cut mean time to repair in the network operations centre by correlating alarms with topology and change records, retrieving the matching runbook, and requiring engineer approval before any automated remediation touches the network.
| Portfolio attribute | Value |
|---|---|
| Primary persona | NOC engineer |
| Primary aggregate | NetworkIncident |
| Cosmos DB containers | 2 |
| Grounding data | alarms, topology, logs, tickets, runbooks |
| Development estimate | USD 60-220 monthly dev estimate |
| Runtime options | Python FastAPI and C# ASP.NET Core |
Productivity hypothesis
These measures reflect the workflow, evidence, decision boundary, and industry risk for this use case.
| Impact type | Measure | Conservative internal assumption | Customer measurement |
|---|---|---|---|
| Productivity | Alarm-to-approved-action time | 10-18% reduction | (baseline median alarm-to-action minutes - assisted median minutes) / baseline median minutes * 100 |
| Quality or risk | Rework rate | 5-10% reduction | (baseline reopened, corrected, or repeated task rate - assisted rate) / baseline rate * 100 |
Important: Conservative internal pilot assumptions only; not public benchmarks, commitments, or expected customer results. Validate against a comparable baseline or matched control cohort with mandatory human review.
Rationale: Network operators can prepare diagnoses and response options faster within existing change controls.
Pilot guardrail: Do not execute network changes autonomously; require operator verification of alarms, topology, logs, tickets, and runbooks.
Customer validation plan
- Baseline the current workflow with representative users and tasks.
- Run the accelerator on the same task mix with human review retained.
- Measure alarm-to-approved-action time and rework rate with the documented formulas.
- Replace internal assumptions with customer-observed results before a business case.
Why Azure Cosmos DB
This AI solution needs an operational data plane, not only a vector index:
- Hybrid vector and full-text retrieval combines semantic recall with exact identifiers, codes, names, and domain language.
- Hierarchical partition keys and metadata predicates keep tenant and domain boundaries in the normal query path.
- Operational JSON, source chunks, workflow and agent state, feedback, and audit records remain together for low-latency reads and governed writes.
- Change feed reacts to source and operational updates without polling.
- Autoscale throughput, configurable consistency, and multi-region distribution support variable demand and enterprise availability.
Domain workflow

| Stage | Actor | Input → output | Approval |
|---|---|---|---|
| 1. Ingest and deduplicate alarms | Fault management integration | Alarm stream from network elements → Deduplicated alarm set per region | Policy controlled |
| 2. Correlate alarms into a service incident | Correlation agent | Alarms plus topology dependencies → NetworkIncident with service impact | Policy controlled |
| 3. Test change and maintenance causation | Causation agent | Recent change records in the blast radius → Suspected cause with supporting change evidence | Policy controlled |
| 4. Retrieve and adapt the runbook | Runbook retrieval agent | Fault signature and element type → Ordered diagnostic and remediation steps | Policy controlled |
| 5. Approve and execute remediation | NOC engineer | RemediationProposal with predicted impact → Executed change with rollback plan and incident update | Human required |
Technical architecture

Cosmos DB data model
| Container | Hierarchical partition key | Documents | Search paths |
|---|---|---|---|
networkIncidents | /tenantId, /networkRegionId | NetworkIncident, NetworkAlarm, TopologyRelation, ChangeRecord, RemediationProposal | /impactSummaryVectorEmbedding, /impactSummaryVector |
networkRunbookLibrary | /tenantId, /faultSignatureCode | RunbookProcedure | /stepVectorEmbedding, /stepVector |
Query contract
| Query | Container | Scope | Purpose |
|---|---|---|---|
| Incident correlation context | networkIncidents | single-partition | Read alarms and topology for the region inside the correlation window in one partition query. |
| Recent change blast radius | networkIncidents | single-partition | Find changes touching the impacted elements shortly before the fault. |
| Runbook signature match | networkRunbookLibrary | single-partition | Knowledge-container query against networkRunbookLibrary (partition-aligned to /tenantId + /faultSignatureCode): retrieve the runbook steps whose fault signature matches the correlated incident; the library stores one runbook document per fault signature so the lookup is a single partition read. |
| Priority incident board | networkIncidents | single-partition | Show unresolved incidents ordered by customer-impacting priority. |
Python and C# services route each document type to its declared container and hierarchical partition key. ETag-based writes protect mutable aggregates, while append-only records and bounded cross-partition queries preserve scale.
Azure deployment

Clone the source repository, authenticate with Azure, and run:
Set-Location accelerators/telecommunications/network-operations-copilot
az login
./Deploy.ps1 -Runtime python -Location eastus2
Change python to csharp to deploy the .NET implementation.
Explore the implementation
- Launch signed-in demo
- Implementation article
- Domain data model
- Workflow and API contract
- Python application
- C# application
- Domain container Bicep
- Platform Bicep
- Cost and configuration
- Source repository