Farm Agronomy Copilot
Outcome
Help the agronomist turn per-field agronomic history, soil tests, in-season imagery and weather into a defensible field-level recommendation for nutrition, irrigation or crop protection, with label restrictions and re-entry intervals enforced before any prescription is written.
| Portfolio attribute | Value |
|---|---|
| Primary persona | Farmer or agronomist |
| Primary aggregate | FieldSeasonRecord |
| Cosmos DB containers | 2 |
| Grounding data | field records, weather, soil, imagery, guidance |
| 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 | Task cycle time | 10-20% reduction | (baseline median task minutes - assisted median task minutes) / baseline median task minutes * 100 |
| Quality or risk | Evidence discovery time | 15-25% reduction | (baseline median evidence-search minutes - assisted median evidence-search minutes) / baseline median evidence-search minutes * 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: Farmers and agronomists can assemble field records and guidance faster, but local conditions require professional judgment.
Pilot guardrail: Keep recommendations advisory and disclose weather, soil, imagery, and local-context freshness limitations.
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 task cycle time and evidence discovery time 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. Field condition sensing | Imagery, weather and sensor connectors | ScoutingObservation → fieldConditionSnapshot | Policy controlled |
| 2. Agronomic diagnosis | Farm agronomy copilot | fieldConditionSnapshot → diagnosisFinding | Policy controlled |
| 3. Label and threshold grounding | Retrieval orchestrator | diagnosisFinding → AgronomicGuidanceChunk | Policy controlled |
| 4. Prescription drafting with restriction checks | Farm agronomy copilot | AgronomicGuidanceChunk → PrescriptionDraft | Policy controlled |
| 5. Agronomist approval and machine release | Certified agronomist or grower | PrescriptionDraft → ApplicationRecord | Human required |
Technical architecture

Cosmos DB data model
| Container | Hierarchical partition key | Documents | Search paths |
|---|---|---|---|
fieldSeasonRecords | /tenantId, /fieldId | Field, SoilTestResult, ScoutingObservation, PrescriptionDraft, ApplicationRecord | /rationaleVectorEmbedding, /rationaleVector |
agronomicLabelLibrary | /tenantId, /cropCode | AgronomicGuidanceChunk | /chunkVectorEmbedding, /chunkVector |
Query contract
| Query | Container | Scope | Purpose |
|---|---|---|---|
| FieldsAtThresholdThisWeek | fieldSeasonRecords | cross-partition-bounded | List fields whose scouting severity crossed the action threshold for the current growth stage. Intentional cross-partition scouting roll-up: bounded by the tenantId prefix, a one-week observation window and the watch-list pest codes. |
| MultiYearSoilTrend | fieldSeasonRecords | single-partition | Compare nutrient levels by zone across seasons to justify a variable-rate nutrition plan. |
| LabelAndThresholdRetrieval | agronomicLabelLibrary | single-partition | Retrieve product label restrictions and extension thresholds for the crop, pest and region in play. |
| PendingPrescriptionsByWindow | fieldSeasonRecords | cross-partition-bounded | Show drafted prescriptions whose application window opens soonest and still await approval. Intentional cross-partition approval queue: bounded by the tenantId prefix, the pending approval state and the open application window. |
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/agriculture-food/farm-agronomy-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