Simplified tedious process using digital approach with state of the art machine learning


KOMINFO (Kementrian Komunikasi dan Informatika) is a ministry of the government of Indonesia that is responsible for communication, information affairs and internet censorship.

One of their most important responsibilities is to make sure that every device that can emit radio signals must have a certificate signed by KOMINFO. This is crucial because an illegal device’s radio signal can interfere with other signals. They can mess-up communication between pilots and ATC (Air Traffic Control) for instance.

To minimize such issues, the team needs to periodically check such products in every digital marketplace in Indonesia by comparing the product name with the e-certificate database manually. If found, they will send a letter to the respective marketplace to ask the seller to remove their product immediately.

For the offline part, the team must swap each of electronic stores across Indonesia, gather the suspicious products and check against the e-certificate database. If found, they will produce a letter and at the later stage, seize the products.

The Brief

Doing all checking manually takes time and is prone to error, for online teams, they struggle to keep up with the number of illegal devices showing up everyday on the marketplace. For offline teams, the process after they identify illegal products till the first letter comes up can take days.  

We were briefed to provide a system that can help them to identify illegal products in an efficient and accurate way for both offline and online teams.


Online Team Challenge
The system must be able to scrape products on marketplaces and check them against e-certificate as fast and yet accurately as possible. There are several challenges we need to address:

We don’t want products from irrelevant categories being checked with e-certificate, e.g. Xiaomi Redmi 9T Pro Tempered Glass will be marked as invalid even though it is tempered glass.

Title Matching
Product’s title from the marketplace will be different from the registered one in e-certificate. How the system know if Jual Xiaomi Redmi 9T Pro 8/16/32 GB Garansi Resmi is the same as Redmi 9T Pro registered in e-certificate.

Offline Team Challenge
The team spread across Indonesia where the Internet signal strength is one of the first issues mentioned to us. The app must be lightweight, work in offline mode and can send and receive data as small as possible. The app must also works both for Android and iOS.

We were only given 2 months to make all the things possible.


For an online team, we create 2 layers of approaches. One is to build a scrape system which can go through online marketplace safely and validate the product’s title against e-certificate using a set of algorithms. The output is a list of products marked with legal, illegal or suspect. The output of this first approach will be used in the second layer.

The 2nd layer is to build a state of the art machine learning that can learn and identify the product. We are using supervised learning using the output from the first layer, setting the correct label if necessary and feeding it to the AI system. By using A.I, we can achieve 85 to 90% accuracy.

Using a combination of algorithm and A.I achieve the online team goals. Process that usually takes weeks to complete, now can be done in 1 day. That is 85% efficiency!

Our approach for the offline team is to build an native apps that can work for both Android and iOS. Due to the timeline constraint, we chose Flutter Framework. Flutter can deliver native experience, yet only need half time of manpowers vs full native. That’s 50% less development time! We build all functionalities needed for more efficient and faster validation process for the team, such as:

 – QR Scan to check products certificate,
 – Digital signature
 – Real time PDF Generation for letter creation
 – With such functionalities, the team can deliver letters to the shop owner in real time!


We also build a caching strategy so the data is safe when the network signal is unstable.

Key Features

Machine Learning
Validation Algorithm
Digital Signature
PDF Generation

Standard Features

Oauth 2.0
W3C AA Standard
OWASP Standard