Moosa Memon
← Index of work

Farming advice that knows when to hand you to a person

Rootstock: a farmer describes what they see; the agent structures it, pulls their soil profile and recent weather, retrieves vetted guidance filtered to their situation, writes an answer where every sentence cites a numbered claim, and then decides whether it should be answering at all.

Type
Agents
Stack
Python / Retrieval / LLM / Weather + soil APIs / CLI
Status
Complete; runs with no API key and no network, both are upgrades

Problem

General agronomic knowledge is abundant. A smallholder’s problem is specific: this soil, this week’s rain, these symptoms on these leaves. Grafting the general onto the specific is the job, and the dangerous part is confident advice on a diagnosis the system isn’t sure about, for a crop someone’s income depends on.

System

rootstock ask "long straight grey marks on the lower leaves of my maize" --location "Kitale, Kenya" runs the pipeline: structure the description, fetch the soil profile and recent weather for the location, compute what those conditions actually imply (not just fetch them), retrieve guidance filtered to crop and conditions, write an answer where every sentence cites a specific numbered claim, then run a final decision: answer, or hand off to an extension officer.

Worth knowing

The hand-off decision is the point of the project. Most advisory agents optimize for always having an answer. Rootstock treats “possibly grey leaf spot, but northern leaf blight presents similarly and I can’t distinguish them from a description; here’s how to tell, and here’s who to call” as a successful output, because for the farmer it is one.

Have a workflow that looks like this?

Most of these start as a messy, manual process someone got tired of. Twenty minutes is usually enough to sketch how I'd approach yours.