Skip to main content
OIL-01 · Oil, Gas & Mining

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 attributeValue
Primary personaOperations engineer
Primary aggregateOperationsQuery
Cosmos DB containers2
Grounding dataprocedures, diagrams, manuals, shift logs
Development estimateUSD 60-220 monthly dev estimate
Runtime optionsPython FastAPI and C# ASP.NET Core

Productivity hypothesis​

These measures reflect the workflow, evidence, decision boundary, and industry risk for this use case.

Impact typeMeasureConservative internal assumptionCustomer measurement
ProductivityEvidence discovery time15-25% reduction(baseline median evidence-search minutes - assisted median evidence-search minutes) / baseline median evidence-search minutes * 100
Quality or riskRework rate5-10% 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​

  1. Baseline the current workflow with representative users and tasks.
  2. Run the accelerator on the same task mix with human review retained.
  3. Measure evidence discovery time and rework rate with the documented formulas.
  4. 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​

Domain workflow

StageActorInput → outputApproval
1. Ask an operations question in contextOperations engineerQuestion with equipment tag and operating mode → OperationsQuery bound to the facility and areaPolicy controlled
2. Retrieve procedures, diagrams, and tagsRetrieval agentQuestion, tag, and current revision filters → Current-revision evidence setPolicy controlled
3. Correlate recent shift historyShift context agentShift log entries referencing the tags → Recent changes and abnormal conditionsPolicy controlled
4. Answer within the operating envelopeOperations copilotProcedure limits and equipment design data → Cited answer stating the applicable limits and authorityPolicy controlled
5. Approve any deviation from limitsArea authority or operations superintendentDeviationRequest with risk assessment → Approved or refused deviation with compensating controlsHuman required

Technical architecture​

Technical architecture

Cosmos DB data model​

ContainerHierarchical partition keyDocumentsSearch paths
operationsQueries/tenantId, /facilityIdOperationsQuery, ProcessDiagramAsset, EquipmentTagRecord, ShiftLogEntry, DeviationRequest/questionVectorEmbedding, /diagramSummaryVectorEmbedding, /entryVectorEmbedding, /questionVector, /diagramSummaryVector, /entryVector
operatingProcedureLibrary/tenantId, /procedureIdOperatingProcedure/stepVectorEmbedding, /stepVector

Query contract​

QueryContainerScopePurpose
Tag-centric context loadoperationsQueriessingle-partitionRead equipment master, related diagrams, and recent shift entries for a tag in one partition query.
Current-revision procedure retrievaloperatingProcedureLibrarysingle-partitionKnowledge-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 conditionsoperationsQueriessingle-partitionShow shift log entries in the area that flagged abnormal operation.
Open deviation registeroperationsQueriessingle-partitionList 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​

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​

Return to all accelerators