Agriculture · Vision AI
Krishi Lens
Point a phone at a sick crop. Get a diagnosis, a treatment and a price — in your language.
A voice-and-camera farm advisor for smallholders. It diagnoses pests and disease from a photo, recommends the right treatment and dose, warns about the weather, and tells farmers when and where to sell. Free for farmers; paid for by the input dealers, co-ops, lenders and insurers who need farmers to succeed.
Launch market: Bangladesh · 4 cited sources · Facts reviewed 14 October 2026
- Cash to MVP (est.)Our estimate of the cash needed to reach a working MVP: one-time setup plus monthly running costs for the build period. It excludes founder salaries.
- $4,580
- MVP timelineMinimum viable product: the smallest version real users will use and pay for, built to learn fast.
- 8 weeks
- Margin after AI (est.)Of every 100 you charge, how much is left after paying for the AI models and infrastructure behind it. Higher means a healthier business.
- ~80%
- Editorial scorePlankito's own judgement: the average of six 1–5 ratings (AI leverage, moat, competition gap, capital efficiency, timing, scalability). It is an opinion to compare ideas, not a forecast.
- 4.3 / 5
At a glance
- Who uses it
- Smallholder farmers. It is free for them.
- Who pays
- Input dealers, co-ops, lenders, insurers and farm programmes.
- Who should build it
- An ML engineer plus an agronomist with access to real fields.
- The hardest part
- Diagnosis accuracy. A wrong answer can cost a harvest.
01
Opportunity
Smallholders lose crops to pests and disease they cannot identify in time, overuse the wrong chemicals on a shop-keeper's guess, and sell at the worst moment. Government extension officers are few and far between. The advice that exists is in the wrong language, the wrong format, or arrives too late.
The evidence
⅓
Globalof the world's food is produced by small family farms of under two hectares.
13.3%
Globalof the world's food — 1.31 billion tonnes — is lost between harvest and retail every year.
95%+
Globalof the world's languages are low-resource, which is why most farm AI does not speak to the farmers who need it.
16% vs 29%
Globalgenerative AI use in the Global South versus the Global North — the farmers growing a third of the world's food are on the wrong side.
Why now
- Vision models can now identify common crop diseases from an ordinary phone photo.
- Voice interfaces let farmers who cannot read comfortably ask questions and hear answers.
- Cheap inference means a diagnosis costs a fraction of a cent, so the service can be free to farmers.
- Lenders and insurers increasingly want farm-level data to price risk — and will pay for it with consent.
02
The AI case
What AI makes possible here, why the big labs won't absorb it, and whether the margins hold.
AI leverage — what is newly possible
- Diagnoses pest and disease from a photo, with a confidence level and a second-opinion route to an agronomist.
- Gives treatment, dose and safety advice by voice, in the farmer's language and dialect.
- Combines local weather forecasts with crop stage to warn before a disease outbreak, not after.
- Answers “should I sell now?” using recent local market prices.
Lab-proof moatsA reason competitors, including big AI companies, can't easily copy you: data, distribution, trust or local knowledge.
Labelled field images
Hundreds of thousands of local crop photos labelled by agronomists — local varieties, local diseases, local light. Generic vision models lack them.
Dealer and co-op distribution
Input dealers and co-ops sign up whole villages. That channel takes years to build and is invisible to global apps.
Outcome data
Which advice led to which yield. Over seasons this becomes a proprietary agronomic model and a credit signal.
Inference economicsThe cost of running AI models each time a customer uses the product. It decides whether an AI business makes money. (estimates)
- Unit
- Farmer per season
- AI cost per unit
- ~$0.20
- Price per unit
- ~$1.00–2.00 (paid by dealers, co-ops, lenders or insurers)
- Gross margin
- ~80%
How the margin is protected
- →Run a small on-device or edge model for common diseases; call the cloud only for hard cases.
- →Pre-compute weather and price alerts per district, not per farmer.
- →Sell consented, aggregated insights to lenders and insurers.
03
Market & expansion
Launch in Bangladesh. Then expand.
- Farmers will not pay upfront; the businesses around them will. Sell to dealers, co-ops, microfinance lenders and insurers.
- Start with one high-value crop in one region (e.g. vegetables or rice in a single division) to build accurate data fast.
- Agronomist review of uncertain diagnoses is essential for trust in the first seasons.
- Validate: how often farmers lose a crop to misdiagnosis, and what dealers would pay to reduce product returns and complaints.
Editorial scorecardPlankito's own judgement: the average of six 1–5 ratings (AI leverage, moat, competition gap, capital efficiency, timing, scalability). It is an opinion to compare ideas, not a forecast.
- AI leverage
- 5/5
- Moat durability
- 4/5
- Competition gap
- 4/5
- Capital efficiency
- 4/5
- Timing
- 4/5
- Scalability
- 5/5
Where it travels next
India & Nepal
Similar crops, similar smallholder structures, enormous scale.
Sub-Saharan Africa
The world's highest post-harvest losses and fast mobile adoption.
Southeast Asia
Rice, vegetable and fruit smallholders with strong co-op networks.
04
Competition & edge
Who already plays here, what they leave open, and what could go wrong.
Plant diagnosis apps (e.g. Plantix)
Photo-based crop disease identification.
GapGeneric advice; limited local language voice support and limited local market and dealer integration.
Non-profit AI advisors (e.g. Digital Green's Farmer.Chat)
AI chat advice for farmers, often grant-funded.
GapBuilt around programmes and grants; room for a commercial model tied to dealers and lenders.
Government extension services
Officers and helplines that advise farmers.
GapToo few officers per farmer; slow to reach a field when a disease spreads.
Where the edge comes from
- Free to the farmer, paid by the ecosystem — removes the biggest adoption barrier.
- Voice-first in local dialects, so literacy is not a requirement.
- Every season makes the model more accurate for this region than any global product.
Risks and how to handle them
A wrong diagnosis damages a harvest.
Show confidence; route low-confidence cases to an agronomist; never recommend banned or unsafe chemicals.
Dealers push their own products through the advice.
Keep recommendations product-neutral by active ingredient; disclose sponsorships clearly.
Seasonal usage and slow revenue.
Cover multiple crops and seasons; add lender and insurer revenue that runs year-round.
05
Business model
Dealer & co-op plans
Annual fee per enrolled farmer, with a dealer dashboard.
Risk insights
Consented, aggregated farm data for lenders and insurers.
Programme contracts
Development agencies and governments pay to reach farmers at scale.
06
Stack & data
Modern, proven tools a small team can run without a DevOps hire.
Farmer app
Lightweight Android PWA + WhatsApp + IVR (phone call) access
Reach smartphones, basic phones and shared devices.
Vision
Fine-tuned open-weight vision model, with a multimodal frontier model as fallback
Low cost for common cases; high accuracy for rare ones.
Advisor
Claude API for reasoning over diagnosis, weather and crop stage
Clear, safe, step-by-step advice.
Speech
Speech-to-text and text-to-speech in local dialects
Voice-first for farmers who prefer not to read.
Dashboard
Next.js 16 + Tailwind CSS v4
Dealer, co-op and agronomist consoles.
Database
PostgreSQL + PostGIS + Drizzle ORM
Field locations, outbreaks and district-level alerts.
Data
Weather and market price feeds, refreshed daily
Timely alerts and selling advice.
Hosting & ops
Vercel + GPU inference endpoint + Sentry + PostHog
Simple operations with a dedicated vision endpoint.
Data model6 core tables, for the engineer starting the first migration.
farmers
Farmers and their plots.
- id
- phone
- language
- district
- dealer_id
- consent_flags
plots
Fields and crops.
- id
- farmer_id
- location (geography)
- crop
- sowing_date
- area
diagnoses
A photo-based assessment.
- id
- plot_id
- image_url
- label
- confidence
- reviewed_by
- created_at
advice
What was recommended.
- id
- diagnosis_id
- treatment
- dose
- language
- delivered_via
outcomes
What happened next.
- id
- plot_id
- season
- yield
- loss_reported
- notes
alerts
Weather and outbreak warnings.
- id
- district
- crop
- type
- message
- sent_at
07
Costing
Directional monthly estimates in USD. Your numbers will differ — the structure is what matters.
| Item | MVP / monthWhat the product costs to run each month (hosting, AI models, tools) at MVP stage. | Scale / month |
|---|---|---|
| Hosting & database | $50 | $500 |
| Vision inference (GPU) | $80 | $1,500 |
| Advisor model inference | $50 | $800 |
| Speech & IVR | $60 | $1,200 |
| Agronomist reviewers (part-time) | $400 | $2,500 |
| Field data collection | $200 | $1,500 |
| Total | $840 | $8,000 |
One-time setup
08
MVP & roadmap
Shippable in about 8 weeks, then a path to production.
Build
- One crop, the ten most common pests and diseases.
- Photo diagnosis with confidence and agronomist review for uncertain cases.
- Voice advice in the local language via WhatsApp.
- District weather alerts for that crop.
- Dealer dashboard showing enrolled farmers and outbreaks.
Not yet
- Market price advice
- IVR for basic phones
- Lender data products
- More crops
Roadmap to production
Validate
Weeks 0–4- Collect and label 3,000 field photos with an agronomist
- Sign two input dealers for a pilot
- Benchmark existing vision models on local photos
MVP
Weeks 5–12- 1,000 enrolled farmers
- Diagnosis accuracy above 90% on the test set
- Dealer pays for season two
Traction
Months 4–12- Three crops
- IVR access
- First lender or insurer data partnership
Production
Months 13–24- Second country
- Outcome-based agronomic model
- Seed or programme funding
09
Start this week
Your first steps. Tick them off as you go.
Metrics that prove it is working
- Diagnoses per farmer per season
- Diagnosis accuracy on the held-out test set
- Reported yield change for advised plots
- Revenue per enrolled farmer
10
Sources
Every figure on this page links to its source. Facts reviewed 14 October 2026. Costs and market notes without a citation are our estimates.
- 1.FAO — Small family farmers produce a third of the world's food (2021)
- 2.FAO — International Day of Awareness of Food Loss and Waste (2023 data) (2025)
- 3.Stanford HAI — Mind the (Language) Gap: LLM development in low-resource language contexts (2025)
- 4.Microsoft — The continued state of global AI diffusion in 2026 (2026)