Operations Knowledge Copilot
Outcome
Give operations engineers on a facility or mine site instant, cited access to operating procedures, P&ID diagrams, and shift history so process decisions follow the approved envelope, with area-authority approval for any deviation.
| Portfolio attribute | Value |
|---|---|
| Primary persona | Operations engineer |
| Primary aggregate | OperationsQuery |
| Cosmos DB containers | 2 |
| Grounding data | procedures, diagrams, manuals, shift logs |
| 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 | Evidence discovery time | 15-25% reduction | (baseline median evidence-search minutes - assisted median evidence-search minutes) / baseline median evidence-search 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: Operations engineers can locate and assemble applicable operating knowledge faster without automating field actions.
Pilot guardrail: Use authoritative procedures, diagrams, manuals, and shift logs; keep operational decisions human-controlled.
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 evidence discovery 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. Ask an operations question in context | Operations engineer | Question with equipment tag and operating mode → OperationsQuery bound to the facility and area | Policy controlled |
| 2. Retrieve procedures, diagrams, and tags | Retrieval agent | Question, tag, and current revision filters → Current-revision evidence set | Policy controlled |
| 3. Correlate recent shift history | Shift context agent | Shift log entries referencing the tags → Recent changes and abnormal conditions | Policy controlled |
| 4. Answer within the operating envelope | Operations copilot | Procedure limits and equipment design data → Cited answer stating the applicable limits and authority | Policy controlled |
| 5. Approve any deviation from limits | Area authority or operations superintendent | DeviationRequest with risk assessment → Approved or refused deviation with compensating controls | Human required |
Technical architecture

Cosmos DB data model
| Container | Hierarchical partition key | Documents | Search paths |
|---|---|---|---|
operationsQueries | /tenantId, /facilityId | OperationsQuery, ProcessDiagramAsset, EquipmentTagRecord, ShiftLogEntry, DeviationRequest | /questionVectorEmbedding, /diagramSummaryVectorEmbedding, /entryVectorEmbedding, /questionVector, /diagramSummaryVector, /entryVector |
operatingProcedureLibrary | /tenantId, /procedureId | OperatingProcedure | /stepVectorEmbedding, /stepVector |
Query contract
| Query | Container | Scope | Purpose |
|---|---|---|---|
| Tag-centric context load | operationsQueries | single-partition | Read equipment master, related diagrams, and recent shift entries for a tag in one partition query. |
| Current-revision procedure retrieval | operatingProcedureLibrary | single-partition | Knowledge-container query against operatingProcedureLibrary (partition-aligned to /tenantId + /procedureId): return only the current approved revision of the operating procedure steps, read as a single partition query on the procedure. |
| Recent abnormal conditions | operationsQueries | single-partition | Show shift log entries in the area that flagged abnormal operation. |
| Open deviation register | operationsQueries | single-partition | List deviations awaiting authority decision or still active with compensating controls. |
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/oil-gas-mining/operations-knowledge-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