Contextual AI for Rwanda

Teaching AI systems to understand Rwandan places, culture, and systems the way people actually describe them.

Elyssa Irankunda

Started November 10, 2025

This research is active and ongoing. We'll publish findings here as they're validated, rather than in advance of them.

"Near the market, past the blue church" is a perfectly good location to a Rwandan speaker and useless to most AI systems, which expect a formal address or GPS coordinates. This project works on the understanding layer that lets our products — and eventually, others' — understand Rwanda the way people actually talk about it.

What we're working on

Understanding that combines formal administrative data (Districts, Sectors) with informal, commonly-used landmark references

Cultural modeling of Rwandan names, communication patterns, and everyday practices

Understanding of how agriculture, healthcare, education, and business actually operate locally

Matching that connects the same concept described differently across languages or phrasings

Abstract illustration of two different paths leading to the same point. — The gap this project addresses: two different descriptions of the same place should point to the same place.

Why this matters for almost everything else we build

Every product ATAS ships eventually runs into the same underlying problem: Rwanda, described the way people actually describe it, does not look like a clean database. This is one of the areas we're working to solve well once, rather than having each product team solve a smaller version of it separately.

Interested in collaborating?

Cite this work

ATAS Research Team (2026). Contextual AI for Rwanda: Geographic and Cultural Context Understanding. ATAS Research Center. https://atas.rw/research/contextual-ai-for-rwanda

Project timeline

November 10, 2025
Project kicked off
Began exploring the gap between formal administrative data (Districts, Sectors) and how Rwandans actually describe locations.

October 1, 2026Planned
Extend to sector-level cultural context
Extending this work toward locally-specific naming conventions and communication patterns at the sector level.