← all challenges
// challenge brief · hack a ton 2026
Crop Whisperer
A Dobrogea farmer describes the problem in plain Romanian; the agent gives a real, grounded answer.
- // Proposed by
- Ambasada
- // Industry
- Agriculture
- // Difficulty
- 🟢 Beginner-friendly
// the problem
Expert agronomy advice costs money small farms won't spend, so problems get guessed at. Dobrogea grows wheat, sunflower, rapeseed, and maize, and a farmer staring at spotting leaves usually has no one to ask in the moment.
// your mission
Build a conversational advisor a farmer talks to in plain Romanian ("my sunflower leaves are spotting, it rained yesterday, what do I do?") that reasons over agronomy guidance plus the local weather and gives a concrete, grounded answer, optionally from a photo.
// how it works
- 1Ask: the farmer describes the problem (text, and optionally a photo of the plant).
- 2Reason: the agent combines agronomy guides, the crop, and the local forecast.
- 3Advise: gives a concrete recommendation (what it likely is, what to do, when to act), with a confidence level.
- 4Refer: when unsure, it says so and points to a human agronomist rather than guessing.
// suggested approach
- Inference: for model access, we suggest [LLMok](https://llmok.app). Use code
AMBASADA26for 50% off. - LLM + RAG over agronomy guides you gather, plus a free public weather API call. There is no image dataset to assemble (photo input via a vision model is a bonus, not a requirement).
- Floor version: text-only Q&A grounded in the guides for one crop. Add the weather context, then optional photo input.
// how you'll be judged
- Soundness and groundedness of the advice (cites guidance; doesn't invent chemicals or doses).
- Calibrated uncertainty: confident when it should be, cautious when it shouldn't.
- Usefulness in a real field situation.
- Clarity in plain Romanian.
- Creativity: weather-aware timing, photo input, follow-up reminders.
// stretch goals
- Photo-based symptom input via a vision model.
- Weather-aware spray-timing advice.
- A seasonal checklist tailored to the farm's crops.
// deliverables
- Git repo: code + README + run instructions.
- Demo: real crop questions answered, grounded and weather-aware.
- 5-minute pitch: architecture, decisions, trade-offs.