Questions we are working on

Our research comes from running real products: every question below started with something we saw people, places or models do.

Location-aware, multilingual AI

Local context retrieval

How an assistant should collect current local information (reports, weather, services, announcements, routes) and answer only from it, showing its sources and fetch times.

Kiswahili language quality

How large language models fail in Kiswahili, how to detect failures automatically, and how to route work to the model that handles it best. Read the note

AI agents that prepare, people who decide

Assistants that turn a conversation into a structured report or map entry, checked against allowed values, and always sent by a person.

Speech intelligence

Spoken local briefings in a natural Kiswahili and English voice, with the phone's own voice as a fallback, for people who listen rather than read.

Reliability on a budget

Chains of several AI providers with health tracking and automatic fallback, so answers stay available and affordable.

Responsible AI

Wording rules that never overstate safety, AI-assisted moderation with human review, and transparency about what the AI was given.

Maps that communities make and trust

Privacy-preserving location

Showing where people are without revealing them: fixed offsets, circles instead of points, and hotspots built only from moved positions. Read the note

Community mapping

Residents mapping water points, clinics and shops with photos and opening hours, with review, duplicate checks, trust levels and certificates.

Hotspots and spatial analytics

Finding clusters of problems and activity to show utilities and councils where attention is needed.

GIS integration

Open standards (OGC API Features) so professionals can read and edit Jirani data in QGIS, including utility staff answering reports.

Route-aware alerts

Spoken alerts along a trip: rain ahead, announcements on the route and services ahead.

Geofencing and coverage

Groups, announcements and alerts bound to a place and radius, and how far each person chooses to be found.

From a report to a repair

How citizen reporting, neighbour verification and routing to the right office change how fast public services respond, starting with water services.

AI-supported citizen reporting for water services

A proposed study with an African urban water authority: does AI in a community platform increase participation, are its Kiswahili answers trusted, and does it shorten response times to leaks and interruptions?

Research design

Verification by neighbours

How "I also saw this" confirmations, AI checks and moderator review combine into reports that offices can act on with confidence.

Ongoing