AMR-Edge Rwanda Small AI for Development

An INT8 model on the meter reads the register and sends a 20-byte authenticated reading over cellular to WASAC and the customer's phone.

Loading model
01

Acquisition

Edge node prototype: ESP32 camera module on a 3D-printed meter mount
Edge node prototype: ESP32 camera module on a 3D-printed mount that clips over the meter. Here an uploaded photo stands in for the camera frame.
or drop a JPEG or PNG here
02

Register localisation

Select a meter image to run the pipeline.
Integer wheel (m³) Fraction wheel (red drum) Register window Glyph candidate

03

INT8 inference

Per-wheel classification normalised 24×32 int8 input · softmax over int8 logits

Waiting for wheel regions.

Extracted reading

—
Model: depthwise-separable CNN, full-integer post-training quantisation
OpOutput (H×W×C)MACs
04

Telemetry packet

Payload: 20 bytes, AES-128-CTR + AES-CMAC-32

FieldValueOn-air bytesProtection

MQTT 3.1.1 PUBLISH frame

05

Cellular uplink

Modem UART (AT commands)


          

        

      

    


    

      

        
06
WASAC operations
simulated back end

        

          

            
Account

            

            
Ingest decision

            
—

            
    
          
    
          
    
            
    Reading ledger
    
            
    Time (UTC)Register m³Used m³m³/dayBill RWFStatus
    
          
    
        
    
      
    

      
    
        
    Customer handset
    
        
    
          
    
          
    
            
    MTN RW--:--
    
            
    No messages
    
            
    ReplyBack