Community infrastructure intelligence · Uganda
From a photo on the street to an evidence‑backed incident.
Potholes, damaged roads, blocked drains and flooding get reported in fragments — a photo here, a message there. MtaaSense uses Meta’s multimodal models to turn those fragments into structured, reviewable records, and helps reviewers consolidate repeat reports into a single incident.
Reviewer workspace04/15·00:00:09:13
Mbale, Uganda — Floods, Jul 2022
Footage: Dixon099 · CC BY-SA 4.0Disasters start as street‑level warning signs
The reel above is news and community footage of floods, washed‑out roads, drought and landslides — from Mbale, Douala, Bangui, Zambia, Somalia and beyond, 2017–2024. Long before damage like that makes the news, residents see the signs: a drain that never clears, a road that floods every time it rains. MtaaSense makes those observations count as evidence.
Live from the database
How a report becomes an incident
- 01
A resident reports
Photo, a few words or a voice note, and an optional map pin. The original evidence is stored unchanged.
Muse Voice Transcribe
- 02
Muse Spark structures it
Category, visible conditions, a preliminary severity and what remains unknown — each observation tagged with its source.
Muse Spark
- 03
SAM 3.1 outlines the hazard
A pixel mask over the pothole, blocked drain or standing water, rendered on the original photo.
SAM 3.1
- 04
Reports become one incident
PostGIS finds nearby reports of the same kind; a reviewer compares the evidence and decides whether to link them.
Human review
Evidence first
Original photos, descriptions and timestamps are never overwritten. Every decision is in the audit log.
AI kept distinct
Model output is labelled as interpretation — with its provenance — and never presented as resident fact or official verification.
People decide
Duplicates are suggestions. Only a reviewer links reports, corrects categories or marks an incident resolved.