Development roadmap
Built in the open, labelled honestly.
Statuses are assigned from evidence in the repository - tests, configuration and documented verification. A milestone is never marked complete because a folder or placeholder exists.

Showing 10 of 10 milestones
- 01
Completed
Product discovery and architecture
Product requirements for the mobile app, backend and AI/RAG service, and a three-layer architecture with clear trust boundaries.
Evidence
- - Frontend, Backend and AI-RAG PRDs in the repository
- - Architecture decision records
- 02
Completed
Mobile frontend
The Expo / React Native app: onboarding, plots, crop lifecycle, scan, chat, weather, notifications, search, settings and localization.
Evidence
- - Typed, linted codebase with store-level tests
- - CI workflow and EAS build configuration
Next
Device verification of map drawing and live camera preview; broader component test coverage.
- 03
Completed
Backend and data model
A FastAPI modular monolith with enforced domain boundaries, PostgreSQL + pgvector, Redis, Celery workers, admin tooling and audit logging.
Evidence
- - Unit and integration test suites
- - Import-boundary contracts enforced in CI
- - Docker and Kubernetes deployment manifests
- 04
In progress
AI service and model integration
Vision, OCR, RAG, LangGraph agents, memory, safety and confidence scoring are built. Production-tier models are configured but not deployed on GPU hardware.
Evidence
- - Unit and integration tests against real Postgres and Qdrant
- - Documented known limitations
Next
Object detection, structured-output enforcement and production-model deployment.
- 05
In progress
Agricultural knowledge base
The ingestion and retrieval pipeline works; the corpus is a small seed set. A reviewed, provenance-tracked library of agronomic guidance is still needed.
Evidence
- - Ingestion pipeline with dedup and OCR fallback
- - Golden-dataset retrieval evaluation
Next
Curate extension and research documents with expert input; add region metadata.
- 06
In progress
Safety and evaluation
Safety screening, verification and drift detection exist. The semantic safety check has a measured gap on the development model.
Evidence
- - Safety and evaluation modules with tests
- - Documented limitation from live testing
Next
Re-test on production-tier models; build a labelled safety evaluation set.
- 07
In validation
End-to-end integration testing
Mobile ↔ backend flows are live-verified. The backend still runs its AI client in mock mode by default, so the full app → backend → AI service loop needs validation.
Evidence
- - Live-verified scan flow against the backend
- - Live-verified AI service → backend context call
Next
Run the full loop with real models and record latency and failure modes.
- 08
Planned
Pilot validation with agricultural experts
Expert review of diagnoses and recommendations on representative, locally sourced cases before any farmer-facing use.
Next
Identify collaborating agronomists and extension partners.
- 09
Research direction
Sustainability-impact evaluation
Measure whether decision support changes recommendation patterns - for example, the rate of unnecessary chemical recommendations on a defined dataset.
- 10
Research direction
Field studies
Longer-term studies of real-world decisions, input use and outcomes, with appropriate partners, ethics review and measurement methods.
Milestone statuses are maintained in a single data file (src/content/roadmap.ts) and were last reviewed in October 2026.