PhD, Computer Science — AI
2031 · UC Berkeley
MSc Computer Science, thesis
2028–2030 · McGill
Research intern at Mila
2028 · Structured reasoning in LLMs
Data engineer at Capgemini
2027 · Internship
Data science & cloud at ENSA Oujda
2025–2028 · National scholarship
Software engineer at DGI
2025 · Ouagadougou · Internship
Prep classes at ENSA Oujda
2023–2025 · National scholarship
Systems
Products built under the constraints that make the questions hard.
Low-end devices, unreliable networks, cash-adjacent payment habits, and user bases small enough that every measurement is noisy. These are the conditions, not the excuses.
AfroBite · Site
Shipped, not a prototype
afrobite.app is live and serving a v1.0 download page. The marketing site, the four applications and the backend went out together.
Fifteen seconds, filmed in the kitchen
The promise is stated plainly on the page: every dish is a short clip shot where it was actually cooked, so you taste it before you tap.
One language across five surfaces
Site, customer, restaurant, courier and admin share a design system and a single backend — including the light and dark themes.
AfroBite · Client
Priced in FCFA, not converted
The catalogue, the currency and the categories are local — a 100 FCFA plate of attiéké sits beside a 4 200 FCFA burger, delivered in Ouagadougou.
Four states, one event stream
Picked up, en route, arrived, delivered. The customer's map and the courier's app read the same events, so the two can never disagree.
The courier, one tap away
Call or message straight from the tracking screen — on a thin network a phone call still closes the last hundred metres faster than a chat thread.
AfroBite · Découvertes
A video feed that sells
Restaurants post vertical clips; the dish, its price and an add-to-cart sit on the video itself, so discovery and ordering become one gesture.
Tuned for low-end devices
Video is adapted to the handset and the connection rather than shipped at a single bitrate and hoped for.
Order without an account
Guest checkout is a first-class path. Asking for an email before the first order is how you lose the first order.
AfroBite · Resto
The merchant side, shipped too
A separate application for restaurants — menu, orders and deliveries — with its own authentication and its own dashboard.
Onboarding that survives reality
Logo, address dropped as a GPS pin, WhatsApp number: the fields a Ouagadougou restaurant can actually fill, checked by the team before going live.
Four apps, one backend
Customer, restaurant, courier and admin, sharing 56 cloud functions and more than 35 collections.
AfroBite · Support
Support where users already are
An official WhatsApp assistant — in Burkina Faso, a support channel that needs an app install is a support channel nobody uses.
It does the work, not just the talk
Order status, payment help, complaints and refund requests are handled in the thread, and handed to a human when it cannot answer.
Scoped on purpose
The assistant answers on AfroBite and declines the rest. The boundary is part of the design, not an afterthought.
agriAI
Soil, plant, weather
Three tools in one place: soil analysis, crop-disease diagnosis from a photograph, and a forecast that arrives with agronomic advice attached.
A phone number is the account
Sign-in by SMS on a +226 number. No email, no password — the credentials a Burkinabè farmer actually has.
Team of four, led
Four modules and twenty-eight documented architecture decisions, positioned with a consulting agronomist against national B2B prospects.
agriAI · WhatsApp
Advice, not data
A dropped pin returns seven days of weather already turned into instructions: water stress ahead, irrigate early, mulch if you can.
No install, no data plan
The whole advisory runs inside a WhatsApp thread — reach does not depend on a Play Store download or a smartphone.
An assistant that knows its limits
It answers on crops, soils and their monitoring, and refuses everything else. A narrow scope is what makes the advice trustworthy.
QCM Platform
AI drafts, the teacher decides
Question generation runs on Groq; the instructor edits, publishes and keeps the last word on every item that reaches a student.
Marking and analysis, same screen
Average score per test, recent papers and their state — active or draft — sit beside the results rather than in a separate export.
A stack chosen per service
Next.js for the interface, Spring Boot and FastAPI behind it, PostgreSQL and Firebase for state — each service reasoned about, and failed, on its own.
Blogs & Research
Making decisions from data you don't have enough of.
Each piece below is stated the way I would want to read it: the question, what I did, what it showed, and what it does not show.
Delta Method variance estimation, CUPED and CUPAC for sensitivity, and how to choose a test under skew. The guide I wanted when I started.
- Experimentation
- CUPED
Haphiz Ouarma· 4 min read
Six months of the practitioner literature, synthesised: Delta Method variance estimation, CUPED and CUPAC for sensitivity, and how to choose a test under skew. A survey, not yet an original contribution.
- Delta Method
- CUPED · CUPAC
- Choosing a test
Haphiz OuarmaBuilding for the marketsthe literature skipped, and measuring them anyway.
My work spans data and AI — machine learning, deep learning, and controlled experimentation from a thousand users to a million. The literature assumes unlimited traffic; what I build runs on low-end devices, over thin networks, for user bases counted in thousands.
I am not learning to become the best at it. I am learning to build what lets a generation of Africans create, innovate and export technology of their own. Africa does not need to wait for its future — it has to build it.
My vision
Where many see a market, I see immense scientific and technological potential. Africa should stop being only a consumer of technology and become a continent that invents, publishes, builds and exports its own.
Why Horizon's
Horizon's is the vehicle for that vision — the place where research, entrepreneurship and education meet, and where ideas are turned into durable impact.
Why the Academy
Everything I learn, I want to pass on. Training builders is the most durable way to multiply the work and to raise the next generation of African engineers.
Why I learn daily
Every new skill brings the mission closer: build technology that is useful, share the knowledge behind it, and help Africa become a centre of innovation.
© 2026
