The AI engine
Intelligence you can trace.
ai krishi uses AI where it is strong, and says so where it is not. Every output is tied to evidence, scoped by role, and open to human review.
Architecture
Six layers, one engine
From raw signals at the bottom to the person acting at the top.
Experience
Farmer app, WhatsApp, command-centre dashboards, reports
Orchestration
A router that classifies the request, gathers evidence from the right agents and synthesises one answer
Agents
24 specialised agents covering advisory, crop health, weather and climate, markets and economics, programmes and administration, and documents
Knowledge
A knowledge graph, a curated agronomy and disease library, and retrieval over trusted documents
Signals
Satellite crop-health series, weather history and forecasts, market prices, field photos, land records, uploaded reports
Governance
Role-based access, jurisdiction scoping, explicit-confirmation workflows, usage monitoring
Responsible AI
Principles we do not compromise on
Grounded
Advice cites field data, satellite signals, weather and a curated knowledge base, not model imagination.
Honest about uncertainty
If there is too little data, the system says so. A transparent trend is labelled a trend, not dressed up as a trained model.
Safe by design
No pesticide brands or doses. Scoped guidance, then a referral to a local expert.
Human in the loop
Anything that changes a record is applied only after explicit confirmation.
Least privilege
Each role and jurisdiction sees only what it needs.
Measured
Outcomes are tracked against baselines so value is proven, not asserted.
Coverage
What the agents cover
Advisory
Crop, variety, season and stage guidance in local languages.
Crop health
Photo diagnosis, pest and disease risk, similar cases, yield-trend estimates.
Weather and climate
Forecasts, field weather, spray windows and rainfall history.
Markets and economics
Prices, trends, support-price comparison, margin views.
Programmes and administration
Schemes, insurance, district and state analytics.
Documents and data
Land-record reading and report extraction into structured indicators.
Want to see the evidence behind an answer?
We will walk your team through real outputs and the data behind them.