Small
Lightweight, well-understood machine learning instead of large opaque systems.
Random Forest and gradient-boosting models
Run affordably on modest infrastructure
Feature importance that can be inspected and explained
Rwanda-first · Prototype in development
AI-Powered Flood, Erosion & Landslide Risk Assessment
Helping African communities make safer land and housing decisions through accessible, explainable environmental intelligence.
Starting in Rwanda, where steep hills and two rainy seasons make land risk a daily reality.
The problem
Rwanda’s hilly terrain, intense seasonal rainfall and fast urban growth mean many homes are built on steep slopes, near wetlands or along natural drainage paths. Floods, soil erosion and landslides recur during the rainy seasons — the floods and landslides of May 2023 in the Western and Northern provinces were among the deadliest in recent years.
For a family buying a plot, these risks are often invisible: a dry valley in July can be under water in April, and a stable-looking hillside can fail after weeks of rain.
Houses built in flood paths or on unstable slopes can be damaged or destroyed, putting families at risk.
Roads, bridges, water and power lines are cut by landslides and washed out by floods.
Erosion strips fertile topsoil from farmland, reducing harvests and income.
Buyers — including the diaspora buying remotely — can invest savings in land that later proves unsafe or unbuildable.
Environmental risk information exists, but it is scattered across technical datasets, expensive to interpret, and rarely available in plain language or in Kinyarwanda. Most buyers decide without it.
How it works
Each step is labelled with its current status, so you can see exactly what works today and what is still being built.
Search for a place or tap a point on an interactive map of Rwanda.
Working now
Combine terrain, rainfall, soil, drainage and satellite observations for the selected area.
Data integration not started
Estimate flood, erosion and landslide risk categories using validated, lightweight models.
No model is running yet
Show contributing factors, uncertainty and recommended precautions in Kinyarwanda or English.
Report layout preview in English
Interactive demonstration
Pick any location in Rwanda, then choose an example scenario to see how a future report will explain risk, uncertainty and next steps.
Search text is sent to OpenStreetMap to find places. Your selected location is never stored.
1 · Selected location
No location selected yet. Search for a place or click the map.
3 · Preview
Select a location first.
Our AI approach
We believe environmental risk tools should be small enough to understand, careful enough to trust, and built for the people who use them.
Lightweight, well-understood machine learning instead of large opaque systems.
Random Forest and gradient-boosting models
Run affordably on modest infrastructure
Feature importance that can be inspected and explained
Designed to inform decisions, not replace professional judgement.
Transparent explanations of every result
Uncertainty communicated alongside each risk level
Privacy by design: locations are rounded and not stored
Human oversight and expert review of model outputs
Built for Rwanda’s landscape, institutions and languages.
Rwanda-specific environmental data
Validation against locally recorded flood and landslide events
Kinyarwanda and English explanations
Public benefit & open science
We intend to provide essential environmental risk information free of charge and to build in the open. Our planned commitments:
Publish data pipelines and model code under an open-source licence.
Document how models are built and report validation results — including where they perform poorly.
Share derived environmental datasets where licensing and privacy permit.
Work with researchers, engineers, local institutions and communities on design and validation.
Help buyers, builders and planners avoid high-risk land and plan precautions.
These are commitments for the project going forward; no code, data or validation results have been published yet.
Roadmap
Milestones are proposed and may change as we learn from validation and community feedback.
Phase 1
In progressIntegrate terrain, rainfall, soil and land-cover data; build and test the first models; ship the explanation interface.
Interface prototype is live on this page; data integration and models are not yet built.
Phase 2
ProposedValidate models against recorded events with research partners, and pilot with communities and buyers.
Phase 3
ProposedExtend validated coverage across all of Rwanda’s districts, with Kinyarwanda explanations.
Phase 4
ProposedExplore Uganda and Kenya — only after Rwanda results are validated.
Project team
A Hills Props team combining software and AI engineering, real estate, research and partnerships.

CEO · Senior Software and AI Engineer
Engineering · Software Engineer II at Microsoft
Technology entrepreneurship and social impact
Technology partnerships and business development
Rwanda real estate and economics
Research, evaluation, and product management
Get involved
We are looking for researchers, engineers, data partners and communities to shape and validate Land Safety Check.
Or email us at info@hillsprops.com