Inspiration Farmworkers often see a problem first. We wanted a tool that turns one photo into something a worker can act on and a supervisor can check, in the language they speak. What it does Upload a photo of a crop, and CropSight tells you what it sees, what problems could be behind it, and what to check next. It works in English and Spanish. It also includes a summary a supervisor can copy. It is not a diagnosis, and it says so. How we built it A small Node.js server sends the photo to OpenAI’s GPT-4o mini, checks the reply against a fixed format, and sends back a result. The front end is a simple web page that works on a phone. We wrote crop and seed guides for Central Valley crops and gave them to the model as text. Challenges we ran into Seed identification was the hardest part. Early versions got species wrong or said “Unknown.” We rewrote the seed guide with exact measurements and made the model choose from a fixed list. Accomplishments that we’re proud of It works end to end, in both languages, and it tells you what it can’t tell from the photo. What we learned Checking the AI’s output matters as much as the prompt. Being honest about limits earns more trust than a confident guess. What’s next for CropSight Test seed results against real labeled photos, host the app so anyone can open it, and try it with farmworkers and supervisors in the field.