Field Service Grid Copilot
Outcome
Equip a field technician at the pole or substation with the right asset record, switching procedure, and outage context to complete the work order safely, with a control-room switching authorization before any energized operation.
| Portfolio attribute | Value |
|---|---|
| Primary persona | Field technician |
| Primary aggregate | FieldWorkOrder |
| Cosmos DB containers | 2 |
| Grounding data | asset registry, GIS, manuals, outages, work orders |
| 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 | Field task cycle time | 8-15% reduction | (baseline median field-task minutes - assisted median minutes) / baseline median minutes * 100 |
| Quality or risk | Reopened work orders | 3-6% reduction | (baseline reopened work-order 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: Asset registry, GIS, manuals, outages, and work orders support faster field preparation with tight safety controls.
Pilot guardrail: Safety procedures, asset-system truth, and technician verification take precedence over generated guidance.
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 field task cycle time and reopened work orders 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. Accept the work order in the field | Field technician | Dispatched job with asset and location → FieldWorkOrder accepted with crew identity | Policy controlled |
| 2. Load asset, connectivity, and outage context | Network context service | Asset identifiers and feeder topology → Asset record with upstream devices and active outages | Policy controlled |
| 3. Retrieve the applicable procedure | Retrieval agent | Job type, asset class, and voltage class → Ordered switching steps with safety precautions | Policy controlled |
| 4. Guide the field task with citations | Field copilot | Procedure steps and asset specifics → Step-by-step guidance with manual citations | Policy controlled |
| 5. Authorize switching and energization | Control-room operator | Requested switching operation and grounding state → Issued switching authorization and permit record | Human required |
Technical architecture

Cosmos DB data model
| Container | Hierarchical partition key | Documents | Search paths |
|---|---|---|---|
fieldWorkOrders | /tenantId, /serviceTerritoryId | FieldWorkOrder, GridAssetRecord, OutageIncident | Operational fields only |
gridProcedureLibrary | /tenantId, /documentId | SwitchingProcedure, TechnicalManualChunk | /stepVectorEmbedding, /bodyVectorEmbedding, /stepVector, /bodyVector |
Query contract
| Query | Container | Scope | Purpose |
|---|---|---|---|
| Job context assembly | fieldWorkOrders | single-partition | Read work order, target assets, and active outages for the territory in one partition query. |
| Voltage-correct procedure retrieval | gridProcedureLibrary | single-partition | Knowledge-container query against gridProcedureLibrary (partition-aligned to /tenantId + /documentId): return only the switching steps valid for the asset class and voltage the crew is working on, read as a single partition query on the procedure document. |
| Active outage impact | fieldWorkOrders | single-partition | Show open outages on the feeder the crew is about to work on. |
| Overdue asset inspections | fieldWorkOrders | single-partition | Identify assets in the territory past their inspection interval for opportunistic work. |
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/energy-utilities/field-service-grid-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