Success Stories

FingerSense –

The Story of Qeexo AutoML

Qeexo AutoML was developed as an in-house solution to help Qeexo meet escalating demands, enabling FingerSense to scale to over 300 million devices in four years.

FingerSense is the world’s first software-only solution that can distinguish between touchscreen inputs such as fingertip, knuckle, nail, and stylus. It is powered by Qeexo’s lightweight machine learning technology and leverages data from existing smartphone sensors.


Huawei, the world’s third largest smartphone manufacturer, uses Qeexo FingerSense technology to bring advanced functions to their users. For example, users can use a knuckle to knock twice on the display to take a screenshot, or to draw the letter “M” using a knuckle to open a user-defined music app.


The FingerSense launch on Huawei smartphones was such a huge success that Huawei decided to make it a distinguishing feature of their smartphones. In order to meet the escalating demand of nearly 100 commercial projects per year, Qeexo needed to automate the end-to-end machine learning development process – this automated machine learning platform eventually matured into Qeexo AutoML.

Without Qeexo AutoML:

  • Labor-intensive: Each hardware variant is an individual machine learning problem that required tuning by machine learning engineers.
  • Time-consuming: From data visualization and pre-processing to machine learning model deployment, it required an average of 20 hours for one experienced machine learning engineer to implement one hardware variant.
  • Difficult to scale: Demand for FingerSense grew to nearly 100 commercial projects per year. Without Qeexo AutoML, Qeexo would not have been able to meet this demand.

With Qeexo AutoML:

  • Qeexo AutoML automatically pre-processes data, generates visualizations and performance reports, and builds machine learning models using multiple algorithms.
  • Compared to spending 20 hours to develop manually, Qeexo AutoML allows non-machine-learning-experts to build the same production-ready model within minutes.
  • Machine learning engineers were freed to do more critical and complex R&D tasks, such as performance improvements and optimization research. With this reclaimed time, they enhanced FingerSense’s overall performance by 15%.
  • Qeexo AutoML enabled the deployment of FingerSense on 300+ million devices worldwide in just over four years.

Magic Wand – TinyML, Big Fun

Using Qeexo AutoML, customers can easily enhance product intelligence and rapidly iterate through prototypes to reach a commercial product.

GP is a toy manufacturing company in China that wanted to produce a toy magic wand. Equipped with an accelerometer and gyroscope, the wand would play different sounds when performing various gestures in the air. Unfortunately, the gesture-recognition accuracy was unsatisfactory and the product could not be commercialized prior to Qeexo AutoML.

Without Qeexo AutoML:

  • Lack of knowledge and expertise to process sensor signals and utilize sensor data.
  • Lack of budget to hire qualified machine learning engineers.
  • Rule-based logic had low performance with many breaking cases.

With Qeexo AutoML:

  • Qeexo AutoML automatically extracts features from data and develops optimal embedded machine learning models for GP’s hardware, greatly improving the gesture-recognition accuracy.
  • Qeexo AutoML allows GP to take Magic Wand to production without having to hire a team of machine learning engineers.
  • Machine learning models built with Qeexo AutoML are highly optimized and have an incredibly small memory footprint – ideal for ultra low-power, low-latency applications on MCUs such as the Magic Wand.
  • GP is planning to produce a series of “smart toys” using Qeexo AutoML.

Quality Inspection – Qeexo AutoML Vision

Qeexo AutoML Vision enables machine-vision-based quality inspection to automatically vet for defective parts in a smart automotive factory.

KwangSung Corporation is an automotive parts manufacturer and Tier 1 supplier for Hyundai Motor Company and Kia Motors, with operations in Korea, America, China, and India. It wanted a better way to detect defects in its “quarter garnish” parts, including scratches, gas marks, short shots, and black spots. Before Qeexo AutoML, KwangSung employed workers to visually inspect parts with their bare eyes.

Without Qeexo AutoML Vision:

  • Manual labor is expensive and requires training; employee turnover rate is high.
  • Error rate (unidentified defects) is higher when inspection is performed by humans.
  • Alternatives were extremely expensive and required modifications to the production line.

With Qeexo AutoML Vision:

  • Qeexo AutoML Vision does not require expensive infrastructure or equipment and can be integrated with a small fraction of the cost and effort.
  • Qeexo AutoML Vision can be used by existing KwangSung employees, without complicated training and ramp-up time to operate.
  • Machine learning models generated by Qeexo AutoML Vision run locally and do not require internet connectivity, so the data stays on-premise.
  • KwangSung is rolling out Qeexo AutoML Vision to its other factory sites worldwide after initial evaluation in Alabama.

Try Qeexo AutoML

Register for a free evaluation or other SaaS options.

Try Qeexo AutoML

Register for a free evaluation or other SaaS options.