Fieldwise

Your farm. Your options

Fieldwise helps Irish farmers find relevant grants by describing their farm in their own words. Instead of searching through scheme documents and interpreting eligibility rules, a farmer can type, speak, or request a phone call to get a plain-language starting point.

Inspiration

We started with a simple question: how much support goes unclaimed because finding and understanding it is too difficult?

Agricultural grant information is spread across government pages, detailed documents, and technical criteria. For someone busy running a farm, turning that information into a practical next step can be a project in itself.

Our guiding example was Declan, a suckler farmer in West Cork with wet land beside a river. We wanted him to be able to ask, “What could I do with this land?” and get an understandable answer.

What we built

Fieldwise turns a farm description into:

An editable profile showing what the system understood.
Relevant schemes, with reasons they may apply.
Payment estimates where the rules support a calculation.
Official source links and practical next steps.
Clear distinctions between likely eligibility and conditions that need checking.

We also built a phone flow using Twilio, so a farmer can describe their farm aloud and hear a summary, with the results available on screen.

How we built it

We used React, TypeScript, and Tailwind CSS for the interface, Hono for the API, and SQLite with Drizzle for scheme records and saved data. Shared Zod schemas define the contracts between the frontend and backend.

The advice pipeline has three stages:

Understand: OpenAI extracts a structured farm profile from the farmer’s description.
Check: deterministic TypeScript rules evaluate the profile and calculate supported estimates.
Explain: OpenAI phrases the results in plain English.

Our central design decision was that AI interprets and explains; code performs the eligibility checks and calculations.

We manually encoded scheme facts from official sources and built a recorded-response fallback so the core demonstration can run without fresh AI requests.

What we learned

We learned that making complex information accessible requires more than summarising it. People need to see what was understood, correct mistakes, inspect sources, and recognise uncertainty.

We also learned how valuable shared contracts are when three developers work in parallel with AI coding tools. They let us build the interface, API, and integration work independently while keeping the pieces compatible.

Challenges we faced

The hardest part was representing incomplete information honestly. A short description cannot establish land designation, soil suitability, payment entitlements, or every condition of a scheme. We had to distinguish useful guidance from a definitive decision.

Voice introduced another set of constraints: recognition settings, webhook validation, provider timeouts, and keeping the browser updated while a call progresses. We introduced asynchronous polling so a slow AI response would not hold a phone webhook open.

Finally, we balanced speed with reliability through focused pull requests, automated tests, and production smoke checks.

What’s next

Fieldwise is a prototype, not an official eligibility service. Our next steps are broader scheme coverage, a repeatable process for updating source facts, and testing with farmers to understand which explanations and corrections help them most.