Noor is an ambitious mother growing coffee on the slopes of Letefoho, Timor-Leste. She's capable. But is the market treating her well?
Ermera grows close to half of Timor-Leste's coffee. The buyers who pay properly are in Dili.
“Recent heavy rains have damaged many roads and bridges in the country, disrupting people's access to markets.”
— Asian Development Bank, Timor-LesteNoor has harvested fine Timor-Leste coffee. She needs cash within three days to pay her daughter's school fees.
Timor-Leste's rainy season is notorious. And no truck from Dili will risk the road.
One trader's truck makes it up the hill. He names his price. She can't check it, and there's no one else to ask.
Noor isn't the only one on that road. Her neighbours are carrying coffee baskets too.
More farmers, more bargaining power. Only the quantity went up, yet the price got better. This is cooperation. This is Dure.
of 239 studies found that farmer organisations raised their members' incomes.
Bizikova et al., “A scoping review of the contributions of farmers' organizations to smallholder agriculture”, Nature Food, 2020One farm takes the price it's given. A full truck names its own.
One shipment to where the fair prices are, instead of every family making the trip.
Seed, fertiliser and transport cost less, bought for twenty farms at once.
“Farmers who are already marginalized … require additional support before they are able to benefit.”Bizikova et al., Nature Food, 2020
The benefits are real. Getting there is slow, expensive and exclusive, and it ends in a fight over power.
Dure works just like a cooperative. A simple text from any phone is all it takes to help Noor get a fair price.
A farmer Registry is important and effective, but labour-intensive and expensive to keep. Dure works as a market itself, and every deal leaves a footprint. So the Registry builds itself.
Policy makers and extension officers get accurate, good-quality information on limited resources.
AI is not a silver bullet. Noor knows farming better than Dure, so Dure stays out of how she farms. But when the extension officer visits Noor twice a year, they can make the best of it.
Open the sample Registry → Synthetic data for eight weeks in Letefoho.
Tetum is one of the languages large language models overlook most. Pairing Qwen's small AI with a rules engine, Dure reads Tetum texts close to the level of a large model.
Method. Test set: 60 SMS (44 Tetum, 16 English) written independently by Claude Sonnet, not by us, and frozen before any reader was scored. Exact match: crop, quantity, grade, price and intent must all be right. English subset: 16/16 for Dure. A second set of 80 texts we wrote, including questions: rules engine 98%, with the small model 99%. Yes/no replies: 100% on both. The rules engine scores 78% after our latest update (v2.1). Costs: API list prices; sending 1,000 SMS costs $30–$150, so reading them adds less than 1%. Measured 4 Oct 2026.
Dure books the truck and the pickup point. It can't fix the road, and doesn't pretend to.
Farmers' names and numbers stay with Dure. The ministry's Registry shows pseudonymous IDs and village totals, and only for farmers who said yes.
The rules engine keeps taking offers by text. The small model rechecks them later, and any change goes back to the farmer to confirm.
A rules engine reads every text instantly, with no model to load. Only the texts it isn't sure about go to the small model. If the server or the power fails, the rules engine keeps the pool open.
Thanks for paying attention, and enjoy the demo!