Ministry Of ICT And Innovation: Choosing AI Models Rwanda Can Trust.
- Employer
- Ministry of ICT and Innovation, Rwanda
- Period
- Jun - Sep 2026
Which Model, At What Cost
| Model | Quality | Cost | Latency | Kinyarwanda |
|---|---|---|---|---|
| Model A | ●●● | ●○○ | ●●○ | ●○○ |
| Model B | ●●○ | ●●● | ●●● | ●●○ |
| Model C | ●●● | ●●○ | ●○○ | ●●● |
Picked for the pilot
Built With
- Python
- NLP
- LLM Evaluation
- Model Benchmarking
- Kinyarwanda
Overview
A summer in the Responsible AI Office
Through the Centre for AI Policy and Innovation (CAIPI) and the CMU-Africa internship programme, I joined the Responsible AI Office in the Innovation and Emerging Technologies Department of Rwanda’s Ministry of ICT and Innovation, contributing to the national effort to build responsible, high-impact AI for the public sector in line with the National AI Policy.
Challenge
Picking a model is a policy decision as much as a technical one
A model that serves citizens has to answer well, cost little enough to run at national scale, respond fast enough for a service counter, and work in Kinyarwanda, not only in English. Few published benchmarks measure all four at once, so the choice cannot simply be copied from elsewhere.
Solution
Benchmarks that guide real deployments
I designed and evaluated AI and NLP solutions to support government service delivery and decision-making, and benchmarked machine learning models across quality, cost, latency, and language support, including Kinyarwanda, to guide model selection for real deployments. The work ran across the full delivery path, from discovery and data through model development to deployment, within a small engineering team.
See More Projects
Web Platform
Proqio
Built a sensors management system with the Monitoring Team, mapping sensor locations and alerts so they are easy to read.
View Case Study
Mobile Application
FitCV
Built a job search companion that parses resumes, imports job listings, and tailors CVs and cover letters with AI.
View Case Study
Open-Source API
Media-Master
Built an open-source FastAPI backend that helps people automate content generation, from video transcoding to AI voice.
View Case Study