Skip to main content
OIL-03 · Oil, Gas & Mining

Exploration Insight Copilot

Outcome​

Help geoscientists and mining engineers evaluate a prospect faster by unifying well logs, assay results, historical reports, and geological models into cited interpretations, with technical authority review before drilling or resource decisions.

Portfolio attributeValue
Primary personaGeoscientist or mining engineer
Primary aggregateProspectEvaluation
Cosmos DB containers2
Grounding datalogs, assays, reports, models, history
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
ProductivityAnalysis task cycle time8-15% reduction(baseline median analysis-task minutes - assisted median minutes) / baseline median minutes * 100
Quality or riskFirst-pass acceptance5-10 percentage-point improvementassisted percentage accepted without material rework - baseline percentage accepted without material rework

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: Logs, assays, reports, models, and exploration history support faster evidence synthesis for geoscientists and engineers.

Pilot guardrail: Keep geological interpretation, reserve implications, and investment decisions expert-owned.

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 analysis task cycle time and first-pass acceptance 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. Define the prospect and evaluation scopeGeoscientistLicence area, target formation, and evaluation objective → ProspectEvaluation with data scopePolicy controlled
2. Assemble subsurface and assay datasetsData assembly serviceWell logs, drillhole assays, and survey data → Quality-flagged dataset inventoryPolicy controlled
3. Mine historical reports for prior interpretationsRetrieval agentLegacy reports and competent-person statements → Prior interpretations with reliability ratingsPolicy controlled
4. Synthesize a cited interpretationInterpretation agentLogs, assays, model versions, prior reports → Interpretation with uncertainty and cited evidencePolicy controlled
5. Technical authority reviewChief geologist or competent personInterpretation and model assumptions → Endorsed evaluation supporting a drilling or resource decisionHuman required

Technical architecture​

Technical architecture

Cosmos DB data model​

ContainerHierarchical partition keyDocumentsSearch paths
prospectEvaluations/tenantId, /prospectIdProspectEvaluation, WellLogSummary, AssayResult, GeologicalModelVersion/assumptionVectorEmbedding, /assumptionVector
subsurfaceReportArchive/tenantId, /reportIdHistoricalReport/bodyVectorEmbedding, /bodyVector

Query contract​

QueryContainerScopePurpose
Prospect dataset assemblyprospectEvaluationssingle-partitionRead logs, assays, and model versions for the prospect in one partition query.
Ore-grade interval screenprospectEvaluationssingle-partitionFind assay intervals above the cut-off grade with passing QA/QC status.
Reliable historical interpretationsubsurfaceReportArchivecross-partition-boundedKnowledge-container query against subsurfaceReportArchive (bounded fan-out across the /tenantId + /reportId partitions named by the shortlist): retrieve legacy report content above a minimum reliability rating for citation, over the bounded report ids returned by vector shortlisting of the basin archive.
Model assumption driftprospectEvaluationssingle-partitionCompare model versions whose assumptions changed between builds.

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/exploration-insight-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