ReshamSetu: a Kannada voice copilot for silk-cocoon farmers
Link to open source: https://github.com/Krrish-agrawal/reshamsetu/blob/main/README.md
Cocoon farmers bring their lots to the Ramanagara market without knowing how much of the lot will be accepted or what it might fetch. Most farm apps need reading, typing and English. Farmers have a phone and their own language.
ReshamSetu makes speaking the whole interface. The farmer photographs a Bivoltine cocoon tray, and computer vision (YOLOv8 + EfficientNet) works out the qualified share. They then talk to an assistant in Kannada or English. It reads back the qualified percentage and gives an indicative earning (estimated saleable kg × past Ramanagara auction price). It sets reminders that arrive on the farmer's Telegram, requests an auction slot only after the farmer says yes, and records the real sale so the farmer can see how far the estimate was off.
Built with Agora (speech recognition), Murf (Kannada voice), Claude through OpenRouter (reasoning and tool calls), Node, MySQL and React. Every price is labelled with its source, photo grading is marked experimental, and nothing is presented as a prediction.
This build was uploaded as a hackathon project



