RxScanner and Small AI: An Offline Neural Network That Catches Counterfeit Drugs in Africa

In 2019, in Cape Town, Nigerian startup RxAll's AI counterfeit-drug scanner, RxScanner, failed because of latency to a cloud server in the US. After the failure, the company's engineers built an offline version of the model for a smartphone in just 2 hours — and that version kicked off the Small AI movement.

How RxScanner works

RxScanner is a portable spectrometer that scans a pill with infrared light and sends its molecular profile to an AI model backed by a pharmaceutical database. Within seconds, the system identifies the drug or flags it as counterfeit. The device was already in use in pharmacies in more than a dozen countries, including Ghana, Kenya, Myanmar, and the founder's home country of Nigeria, where counterfeit drugs kill thousands of people every year.

Why cloud AI failed during the demo

That morning, at a hotel in Cape Town, South Africa, the system didn't respond. The spectrometer was connected to a cloud model, but the data center sat 14 000 kilometers from the demo venue, and there wasn't enough bandwidth.

"Our server was in the US, and just getting the result of a single scan took more than five minutes," says RxAll founder Adebayo Alonge.

Alonge immediately told his engineers to compress the model into a compact, low-power version that could run without a network connection. Two hours later, the finished solution saved the demo.

• The demo failed in 2019 at a hotel in Cape Town, South Africa
• The data center hosting the cloud model was in the US, 14 000 km from the venue
• A single scan took more than 5 minutes due to insufficient bandwidth
• RxAll engineers built the offline version of the model in 2 hours
• RxScanner is already used in pharmacies in more than 12 countries, including Ghana, Kenya, Myanmar, and Nigeria

What the switch to a compact model delivered

The failed demo pushed RxAll to build a new version of the device that verifies a pill's authenticity with no broadband internet, no computers, and not even a stable power supply. Alonge became one of the leading advocates of Small AI — compact models that run locally on low-power devices instead of in cloud data centers.

Small AI is the opposite of the giant language models of wealthy countries, with their hyperscale data centers, billion-dollar investments, and debates about AI consciousness. But for millions of people in developing countries, this kind of AI — small and offline — remains the only option available.

An IEEE Spectrum report frames the Small AI movement as a response to the reality across much of the planet: millions of people live where there is no broadband internet, no reliable electricity, and no access to hyperscale data centers — and for them, the only AI that works is the kind that fits in a pocket.

What this means

The RxScanner case shows that while the industry argues over trillions of parameters and hyperscale data centers, what matters most for healthcare in regions without stable connectivity and power is not a model's raw capability but its ability to run autonomously on an ordinary smartphone.

Frequently asked questions

What is RxScanner?

RxScanner is a handheld spectrometer from Nigerian startup RxAll. It scans a pill with infrared light and uses an AI model backed by a pharmaceutical database to determine within seconds whether the drug is genuine or counterfeit. The device already runs in pharmacies in more than 12 countries, including Ghana, Kenya, Myanmar, and Nigeria.

Why was an offline version of the model needed?

Because the cloud version depended on a server in the US, 14 000 km from where it was used — and with a weak connection, a single scan took more than 5 minutes. The offline model, which RxAll engineers built in 2 hours, runs directly on a smartphone without a network connection.