The Sewers of Kisumu Are Talking: How a Kenyan Scientist's AI Bet Could Warn the World of the Next Outbreak

Every day, in a set of unremarkable manholes scattered across Kisumu and Mombasa, someone lowers a container into the flow beneath the street and pulls up a sample of the city's wastewater. It is not glamorous work. But inside that murky liquid is a running, unfiltered record of what is making a whole population sick — the viruses shedding through people who have not yet felt symptoms, the bacteria quietly learning to shrug off the last antibiotics that still work. For three years, Kenyan scientists have been reading that record. Now they are about to hand it to a machine.

This week the International Livestock Research Institute (ILRI) announced that one of its senior scientists, Samuel Oyola, has been awarded a $1.45 million grant — roughly KSh 187 million — from the Gates Foundation to build an artificial-intelligence system that can predict disease outbreaks and detect drug-resistant microbes earlier than any tool currently in use in the region. For a continent that has spent much of the last decade importing its epidemic warnings from laboratories in Atlanta and Geneva, it is a quietly radical proposition: that the early alarm for the next pandemic might be built in Nairobi, trained on Kenyan sewage, and run by African hands.

The idea hiding in the water

The science rests on a simple, almost humble insight. People start shedding pathogens into toilets before they ever present at a hospital, and long before a health ministry connects a cluster of cases into an "outbreak." Wastewater, sampled continuously, captures that signal for an entire catchment at once — tens of thousands of people summarised in a single flask. During the COVID-19 pandemic, cities around the world discovered they could track the virus's rise and fall through sewage days ahead of clinical data. Oyola's team never switched their version off.

According to ILRI, the project — formally titled "Deploying AI Innovation for Bioinformatics and Genomic Epidemiology" — will draw on a longitudinal dataset built from high-frequency monitoring of 30 urban sites, 18 in Kisumu and 12 in Mombasa. Each sample is run through shotgun metagenomic sequencing, a technique that reads essentially all the genetic material present rather than hunting for one suspect at a time. The result is a firehose of data far too large and too noisy for human analysts to comb through in time to matter. That is precisely the gap the AI is meant to fill.

Teaching a machine to see the outbreak coming

The heart of the grant is prediction, not just detection. Oyola and two PhD candidates will fuse the environmental genomic stream with clinical records of infectious disease and antimicrobial resistance, then train models to recognise the faint patterns that precede a surge. The aim, in his words, is to "transform complex genetic data into actionable public health insights that support faster and more informed decision-making."

"Our work combines wastewater surveillance, genomics, epidemiology, artificial intelligence, and public health to develop effective early-warning systems capable of detecting disease threats and antimicrobial resistance in near real time," Oyola said in the ILRI statement announcing the award. The phrase "near real time" is the whole ambition compressed into three words. Traditional surveillance in much of East Africa runs on a lag of weeks — the time it takes for a sick person to reach a clinic, for a sample to travel to a reference lab, for a result to climb the reporting chain. An AI reading the sewers could, in theory, cut that to days.

Why antimicrobial resistance is the quiet emergency

Outbreak prediction grabs the headlines, but the part of this project that may matter most over the long run is its focus on antimicrobial resistance, or AMR — the slow-motion crisis in which the drugs that underpin modern medicine stop working. AMR does not announce itself with a dramatic curve; it seeps in, infection by infection, until a routine surgery or a childhood pneumonia becomes untreatable. It is one of the leading causes of death worldwide, and it falls hardest on low- and middle-income countries where surveillance is thinnest and access to newer antibiotics is poorest.

Wastewater is an unusually honest witness here, too. Resistant bacteria and the genes that ferry resistance between them show up in sewage across an entire population, not just among the few patients whose samples happen to get cultured. Building an AI that can flag a rising resistance signal in Kisumu before it hardens into a clinical wave would give Kenyan hospitals something they have rarely had: warning, and time to act.

A grant that lands in the middle of a bigger argument

Oyola's award does not arrive in a vacuum. Earlier this year the Gates Foundation and OpenAI announced a $50 million "Horizon" initiative to pilot AI tools in African health systems, part of a broader — and contested — push to route the continent's health future through artificial intelligence. Supporters argue it is exactly the kind of leapfrog Africa has pulled off before, most famously when mobile money outran the world's banks. Skeptics counter that pouring AI into systems still short of nurses, reagents and reliable electricity risks building a sophisticated roof over a house with no walls, and warn about who ultimately owns the data and the models.

What makes Oyola's project a useful test case is that it is not a tool parachuted in from abroad. The dataset is Kenyan. The infrastructure has been running in Kenyan cities for years. The scientists are based at ILRI in Nairobi, and the students being trained are the people who will still be here when the grant cycle ends. If African-led AI in health is going to prove itself anywhere, a home-grown early-warning system built on local sewage is a fair place to start.

The diaspora stake in a Nairobi laboratory

For Kenyans watching from Houston, London, Doha or Toronto, a story about wastewater metagenomics might seem far from the daily texture of diaspora life. It is closer than it looks. The global Kenyan scientific community is deeply woven into exactly this kind of work — the researchers in Western universities and pharmaceutical labs who trained abroad and now weigh whether their expertise has anywhere to land back home. Every grant of this scale that stays in Nairobi is an argument, however small, against the one-way flow of talent: proof that a scientist can do frontier work from Kenya rather than only from a bench in Boston.

There is a more direct connection, as well. Outbreaks do not respect borders or passports. The same diaspora families who send money home, who fly back for funerals and weddings, who move between Nairobi and a dozen foreign cities each year, are also the human threads along which pathogens travel. An early-warning system that catches an outbreak in Kisumu a week sooner is, in a real sense, protecting the relatives waiting at arrivals halls half a world away. For once, the technology meant to keep people safe is being built at home — and the rest of the world may end up borrowing it.