IoT & Embedded
Connected devices that survive the real world. We design the hardware. We write the firmware. We pick the right radio for the job.
Octopi Solutions Est. 2020
We design and ship smart systems end‑to‑end. IoT sensors through computer vision through cloud automation. Plus the integration glue between them.
§ 01 / Capabilities
Hardware, software, AI: same room. The awkward parts between them stop being awkward.
Connected devices that survive the real world. We design the hardware. We write the firmware. We pick the right radio for the job.
Pixels into decisions. Number plate recognition. Attribute classification. Anomaly detection. Built on the right model for the job.
Apps, dashboards, APIs that scale. AWS-native. Readable code. A maintenance story you can live with.
Tying systems together. Sensors to dashboards. Devices to people. Data to action. The plumbing that turns parts into a product.
§ 02 / Process
A short, honest path from conversation to deployed product. No black boxes. No surprises. Just regular notes and steady progress.
Step 01
We start with the problem, not the tech. A short call, a clear scope, an honest estimate.
Step 02
Architecture, hardware spec, UX, whatever the project needs. You see the plan before we build a thing.
Step 03
Iterative engineering with weekly progress notes. You can poke at it as it grows.
Step 04
Deployment. Training. Support after launch. We stay until it's actually working.
§ 03 / Selected work
A snapshot of recent projects across IoT, vision, automation. Click any card for the case file.
§ 04 / About
Octopi Solutions was founded in 2020 to do the work most consultancies push to a sub-contractor: the awkward seam between hardware, software, the cloud.
We're a small studio that ships unusually broad systems. A number-plate gate that opens itself. A beacon network that builds attendance reports. A LoRa link replacing a cellular bill. Most of our work lives at the intersection of disciplines. That's exactly why we stay generalists.
Hardware to cloud, one team. Fewer hand-offs, fewer surprises.
Code you can read, weekly progress notes, no black boxes.
Designed for unreliable networks, harsh environments, real budgets.
§ 05 / Contact
Drop us an email about it. We read every message. No funnels. No auto-replies.

Optical Character Recognition traces back to the 1910s, when Emanuel Goldberg built a machine that could read characters and convert them into telegraph code. A century later, OCR is everywhere. From postal sorting to mobile document scanning.
A site in a remote area needed automated gate access. Stationing a guard was costly and impractical. The existing controller required a physical fob the team kept losing.
A camera-side OCR pipeline that identifies vehicle plates against an approved list and triggers the gate locally. The system runs at the edge, so connectivity drops never lock anyone out. A thin cloud companion keeps the access list in sync.

Bluetooth Low Energy was first developed in 1989 by Dr Nils Rydbeck in response to demand for wireless headsets. Today, BLE beacons broadcast small data packets (Major, Minor, UUID, MAC, RSSI, Tx Power) that listening devices can use to triangulate position and presence.
A facility wanted insight into how its staff moved through the building over a working day. Without intrusive cameras. Without manual sign-in.
BLE tags worn by staff. A network of strategically placed scanners that relays packets to an MQTT broker. An AWS Lambda pipeline that ingests, processes, stores the data. Daily, weekly, monthly behaviour reports get generated automatically. Then emailed to management.

LoRa, NB-IoT, Sigfox all sit under the LPWAN umbrella: low power, wide area. LoRa is the newest of the three, released in June 2015. In South Africa it operates on 433.05–434.79 MHz (EU 433) and 865–868.6 MHz (EU 863-870), per the LoRa Alliance.
The client needed real-time notifications across a long distance with no recurring data costs.
A LoRa peer-to-peer link with battery-operated endpoints. A standalone solution that needs almost no maintenance and very little to install. No SIM cards. No cellular bills. No monthly headache.

The Global System for Mobile (GSM) traces back to 1975, when Henry Kieffer suggested a new mobile spectrum at 900 MHz. The first SMS wasn't sent until 1992, by Neil Papworth to Richard Jarvis. Today, a GSM module connected over UART using AT commands is still the most reliable way to send an SMS from a single-board computer.
A farm operation needed to know immediately about changes on the property: doors opening, equipment moving, sensor thresholds being crossed. Cellular data was unreliable. SMS wasn't.
A small fleet of GSM-equipped sensor nodes that fire off context-rich SMS notifications on event. No app to install. No portal to log into. Just a message on the owner's phone the moment something changes.

In the 1960s, Woodrow Wilson Bledsoe built the first facial classification system using a RAND tablet, recording vertical and horizontal coordinates of key features. It is from those manual measurements that modern face attribute systems descend.
The client wanted to tailor in-store advertising based on the broad characteristics of the audience, without uniquely identifying any individual.
An on-device pipeline using AWS Rekognition and custom image processing that classifies attributes (age range, gender, mood) without retaining identity. The current advert is selected to match the room, then forgotten. Privacy by design.

The term "smart home" dates to 1984, coined by the American Association of House Builders. The promise (controlling lights, climate, appliances from a single interface) has slowly become reality over four decades.
Homeowners wanted to automate the parts of the house they actually used most: the lighting and the geyser.
Retrofit Wi-Fi controllers that bring lights and geysers onto the home network, with mobile control and scheduling. Crucially, every original physical switch keeps working. Nothing in the house is worse if the network goes down.