About
Martin Mbalkam Pavtino is a software engineer and AI researcher based in Yaoundé, Cameroon, affiliated with the University of Ngaoundéré. His work applies deep learning to agriculture and health in data-scarce environments, and extends to multi-agent systems, robotics and AI safety.
The longer version
I came to research through engineering, not the other way round. Before I wrote a paper I wrote production software — Java/Spring and Laravel back-ends, React and Flutter front-ends, deployments on AWS and GCP. That background still shapes how I work: I care about pipelines that run reliably, data that is honestly documented, and results that survive someone else re-running them.
The question that pulled me into research was narrow and stubborn: why do agricultural forecasting systems that work well elsewhere perform poorly in Central Africa? Answering it properly meant first mapping what had already been tried, which became a systematic review of deep learning for crop monitoring — 87 studies, 10 architecture families, screened under PRISMA. It continues today with an experimental study testing whether socio-economic variables measurably improve yield predictions in Cameroon, evaluated across six architectures on maize, cassava and cocoa.
Agriculture is where I publish, but it is not the boundary of the work. I am building an IoT health platform with embedded AI for patient follow-up in sub-Saharan Africa — the same underlying problem of making models useful when data is sparse, intermittent and expensive to collect. I also work on collaborative multi-agent systems and robotics, and I take AI safety seriously as a research question rather than as a disclaimer.
Alongside all of this I compete in machine learning competitions — acoustic species identification, subsurface geology prediction, geolocation from synthetic imagery, knowledge tracing on tutoring transcripts. Competitions are where I stress-test architectures under time pressure before I trust them in research work, and where I learn what actually degrades a model rather than what is supposed to.
Outside the lab I have a personal stake in farming and agricultural entrepreneurship, including work on integrated, circular-economy farm designs for Cameroon. Field reality keeps the modelling honest in a way that no validation split can.
At a glance
- Location — Yaoundé, Cameroon
- Affiliation — University of Ngaoundéré, School of Mathematics and Computer Sciences
- Languages — French and English
- Code — github.com/pavtino
- Competitions — kaggle.com/martinmbalkam
- ORCID — [TO COMPLETE]
- Google Scholar — [TO COMPLETE]