Fishcluster has $1m in commitments for AI and underwater robots that see what Nigeria’s fish farmers cannot

Fishcluster has $1m in commitments for AI and underwater robots that see what Nigeria’s fish farmers cannot

He has been doing this for twenty-odd years, on the same stretch of ground outside Mowe, Nigeria, so the routine has long become second nature. He checks the netting. He lifts the bag, and he knows its weight the way you know the weight of your own child.

Then he raps the hollow galvanised frame of the tarpaulin tank twice with his knuckle, a sharp ring the fish have long since learned to answer, and the surface begins to move. The moment he flings the first handful of pellets, the calm water erupts into a violent, boiling frenzy as hundreds of heavy catfish thrash and fight for the feed.

He scoops and throws. The water breaks where the feed lands. He throws again, they keep coming, and he keeps throwing until it looks right.

Then he stops, because it looks right. Or the bag is empty.

There is no science to the method but Sunday can tell you which of his ponds runs hot in March and which one gives him trouble after heavy rain. He can look at a fish and guestimate its weight to within a few grams. What he cannot do, what twenty years has never once given him, is see whether the ones underneath are hungry or already full.

In February, he lost most of a batch across four days. He called it “disease”, because that is what everybody calls it, and he restocked and moved on. He still does not know what happened in that water.

Depending on who you ask, fish feed is roughly 75 percent of what it costs to run a commercial pond in Nigeria.

“One operator even told us it was closer to 87 percent for them,” Samuel Eze, CEO and founder of Fishcluster, recalls.

Make your startup impossible to overlook

Discover the proven system to pitch your startup to the media, and finally get noticed.

Yet, there is so much threat-bearing waste. Whatever the fish do not eat sinks to the bottom, rots, and turns to ammonia. The ammonia kills slowly enough that nobody thinks to blame the feed.

Eze has spent about six months building a company around the sentence Sunday cannot say out loud.

“When the fish move around, you do not see what they do underneath the water,” he tells me on the first of two calls. “It is invisible to anybody, no matter how qualified or experienced you are.”

Fishcluster, which comes out of stealth today, is his answer to that: an AI and robotics system for commercial fish ponds. The pitch is not to say that farmers are careless. It is that the industry’s entire quality-control method is not efficient.

“Every major feed manufacturer in the country employs aquaculture technical experts (ATEs) who travel from pond to pond, manually taking readings and writing numbers down,” he explains.

The key metric they are chasing is the Feed Conversion Ratio (FCR), which has become the working benchmark for a decade or more. In a market where farmers can switch brands at will, it is how a manufacturer proves its feed works. It is also, Eze points out, arrived at entirely by hand.

“Today in the industry, it is assumed, for example, that one kilogram of feed should produce around 1.5 kilograms of catfish,” Eze adds.

So the ATE is the brand defending itself. The problem, however, is that an ATE covers fifty ponds (a hundred at most) on a route, Eze explains, but ponds change faster than routes do.

“You can check now with the manual tool and in the next ten minutes the problem is happening.”

The compounding effect is what makes things expensive. Feed thrown unevenly or excessively (backed by unreliable ATE data) means some fish grow faster than others. In the case of catfish, those that outgrow their batch start eating it. Oxygen drops. Ammonia climbs. “Nothing” announces itself.

“That death is not overnight at all,” Eze says. “It comes quietly, every single day.”

He is careful to highlight that the problem is not a Nigerian one only. From Africa to Asia to Latin America, he argues, fish farmers are fighting the same invisible enemy.

Fishcluster’s solution is three products, each aimed at one part of the problem.

SENTI-100 (pictured above) is an autonomous underwater robot. FORGE-200 is a feeder that holds feed in an enclosure and distributes it across the pond rather than into one corner. FOS 2.4 (Fishcluster OS) is the software deciding what the pond needs and telling the feeder what to do about it.

During our first call, Eze shares his screen and plays footage from inside a pond. Oxygen, pH, temperature, nitrite and ammonia update in something close to real time. The cameras are reading the fish too, size and weight and how the school is moving, and the system is calling the behaviour: active feeding, stress level low.

None of the components are novel. Sensors exist, feeders exist. What is new is the loop.

A sensor reporting ammonia at a given level has told you nothing by itself. Somebody has to have decided in advance that ammonia at that level, in catfish of that weight, in water at that temperature, is the point at which something has to happen. Those decision points are the actual intelligence in the system, and they did not come from the software side.

Eze recruited top researchers out of Nigerian technical universities to set them. Between them they hold PhDs in fish nutrition, pond systems and aquaculture physiology, with fifteen to twenty years each in the field. One of them has presented at the World Aquaculture Society’s global conference. Some of them keep their own ponds. These are not people who normally take a call from a six-month-old startup.

What that buys, in Eze’s account, is a way to blast through the fifty-pond ceiling.

I put it to him that the obvious cheaper fix is better training for ATEs and more accurate tools in their hands, not slapping equally unreliable AI and robotics on top of the problem. He does not accept the premise. According to him, scattered tools cannot solve what he calls an infrastructural problem.

“If they could, they would have by now,” he insists.

What he describes instead is not a replacement. It is triage of sorts. The system flags the pond with the ammonia problem, and the ATE goes to that pond rather than walking the route and hoping to arrive at the exact time and day the problem is happening.

One ATE, he claims, can then oversee a thousand ponds instead of fifty, and that compounds.

All of these describe a system working while somebody is watching it. It does not, however, tell you what happens on say pond number five hundred and fifty-three, probably six months from now, when it’s raining heavily, network is bad, and nobody is watching at all.

But a founder who can show you the thing running is already clearing a bar most never reach.

For now, large-scale fish pond operators.