Skip to content

The platform

One plot, one season, one continuous record.

AgroExpert is a mobile app backed by an API and a separate AI service. It is designed around four workflows that follow a farmer through the season. Each step below carries its real implementation status.

Two women in bright clothes laugh together in a muddy paddy, each holding bundles of rice seedlings.
Women sing and dance during the rice-planting season in Nepal. Photo: Wikimedman (opens in a new tab), CC BY-SA 4.0.

Status legend

Implemented
Built and exercised by automated tests in its own service.
In validation
Built, but not yet validated end to end, on target hardware, or with real field data.
Partial
Some of the intended capability exists; documented gaps remain.
Designed
Specified or configured, but not yet integrated or running.
Planned
On the roadmap. No working implementation yet.
Research direction
An open question to investigate - not committed engineering work.

01Workflow

Quick crop analysis

Photograph a leaf, root, fruit or soil sample and get a possible explanation - with its uncertainty stated plainly.
  1. 01Capture or select photos in the app, with a live previewIn validation
  2. 02Automatic image validation and quality checksImplemented
  3. 03Attach the plot, crop and stage the photo belongs toImplemented
  4. 04Analyze the evidence in a background jobImplemented
  5. 05Possible diagnosis with a confidence tier and alternativesIn validation
  6. 06Explanation of the supporting evidence, with citationsIn validation
  7. 07Evidence-grounded management steps, without product names or dosesImplemented
  8. 08Ask a follow-up question in chat, seeded with the resultImplemented
The app → backend scan flow is live-verified. The backend’s connection to the real AI service is in validation; in development it uses a mock AI client.
Concept preview - an illustration with sample data, not a screenshot of the app.

02Workflow

Persistent plot management

Each plot keeps its own story, so the next decision starts with what already happened there.
  1. 01Create plots with area, location or a drawn boundaryPartial
  2. 02Add a crop and track it through lifecycle stagesImplemented
  3. 03Log irrigation, fertilizer, pesticide, weeding and other activitiesImplemented
  4. 04Revisit previous scans on the crop timelineImplemented
  5. 05Log partial and final harvestsImplemented
  6. 06Archive a crop cycle into a shareable season reportImplemented
  7. 07Plot history informs later diagnoses (memory)In validation
Map boundary drawing needs a real maps key and device testing. Plot sync is currently one-way (device → server).
Neatly prepared vegetable beds with young plants in front of a row of plastic polytunnels.
Vegetable plots and polytunnels in Kaski district, Nepal. Photo: Anil Simkhada (opens in a new tab), CC BY-SA 3.0.

03Workflow

Product and document understanding

Make the small print on fertilizer, pesticide and seed packaging readable - and keep it separate from advice.
  1. 01Extract label text with OCR (English and Hindi)Partial
  2. 02Decode QR codes and barcodes on packagingPartial
  3. 03Structure label fields such as category, crops and warningsPartial
  4. 04Cross-check low-confidence OCR with a vision modelPartial
  5. 05Read soil-test reports and agricultural documentsPlanned
  6. 06Explain label instructions in the farmer’s languagePlanned
  7. 07Verify product registration against an authoritative sourcePlanned
OCR is implemented and tested inside the AI service, but no backend or app feature calls it yet - so it is marked partial end to end.
Two young seedlings with their first true leaves, each growing from a compact plug of dark soil and roots.
Seedling plugs ready for transplanting, with roots binding the growing medium. Photo: Dwight Sipler (opens in a new tab), CC BY 2.0.

04Workflow

Sustainability-aware decision support

Present the full range of responses - prevention, cultural, biological and, when justified, chemical - in context.
  1. 01Ground recommendations in retrieved agronomic guidanceImplemented
  2. 02Bring crop stage, timeline and cached weather into reasoningImplemented
  3. 03Strip fabricated doses and flag unsafe adviceIn validation
  4. 04Order options by an explicit intervention hierarchyPlanned
  5. 05Discover locally available organic resourcesPlanned
  6. 06Weigh affordability and feasibilityResearch direction
Today’s recommendation step is grounded and safety-screened, but an explicit sustainability ranking and local-resource discovery are design goals, not shipped features.
A group of women farmers sit together at the edge of a vegetable field with mustard in flower behind them.
Women farmers in Nepal’s Terai taking part in a programme offering seeds, vegetable-growing advice and wells for irrigation. Photo: UK Department for International Development (DFID) (opens in a new tab), CC BY 2.0.

What the platform will not do

Boundaries are part of the design.

  • - It will not name a specific product or dose. Labels, local regulations and qualified advisers are authoritative.
  • - It will not present a possible diagnosis as certain, or hide low confidence.
  • - It will not replace agricultural extension workers or agronomists; it should help farmers know when to call them.