These Eight Projects Showcase the Power of Machine Learning on the Edge

Sponsored by Infineon, the Getting Edgy with Machine Learning contest showcased innovative edge machine learning applications.

Participants were challenged to use DEEPCRAFT Studio and the PSOC 6 AI Kit in creating ML projects for smart home and industrial applications. They had to build an edge AI model by collecting sensor data leveraging the kit’s onboard sensors, showcase the model working in the real world, and document the solution as a Hackster tutorial.

After reviewing all the submissions, here are the eight projects that emerged as winners.

AI Detection of Cardiac Abnormalities

The heart is the most important muscle in the body, and heart disorder is a leading cause of death, but for various reasons, not everyone keeps it in great condition. More people need regular cardiac tests to mitigate heart risks. Testing often requires bulky equipment in expensive hospitals. So, EdOliver and Victor Altamirano developed a wearable electrocardiograph using dry electrodes and SparkFun’s AD8232 heart rate monitor.

The dry electrodes make the electrocardiograph usable by athletes during activity. Using machine learning, the collected data is organized and presented to carers in a way that translates to actionable insights, addressing the problem of information overload from wearable health devices.

Do keep in mind that this project is for “demonstrative and illustrative purposes only” and should not be used in place of a certified medical device.

AI-Powered Predictive Industrial Machine Maintenance

Predictive maintenance employs data and analytics to mitigate the risk of equipment breakdown and failure, increase equipment lifespan, and reduce the need for replacement machines.

Eslam Fayed’s project is a predictive maintenance model for industrial machines. It uses the to collect real-time vibration data. The model’s training data was collected using PSOC 6’s integrated inertial measurement unit (IMU), trained in DEEPCRAFT Studio, and deployed to the PSOC 6 via ModusToolbox.

Fayed tests his model on a vacuum cleaner, but the project can be adapted to other industrial machines.

SmartListener: Ambient Sound Classifier

Sound classification uses waveform characteristics to group audio signals into predefined categories for applications like voice recognition, music genre detection, and security monitoring.

Mohammed Ali Bedair created the SmartListener to classify sounds in a home environment. The model was trained using audio samples from the ESC-50 (environmental sound classification) dataset, converted to a 16kHz bit rate for compatibility with DEEPCRAFT Studio.

There are five separate classes for sound classification, including baby crying, fire alarm, glass breaking, footsteps, and barks. The device listens for these specific sounds and sends an MQTT alert when they are detected. It uses a TensorFlow Lite model trained in DEEPCRAFT Studio and has a 3D-printed enclosure for deployment in any room.

SmartScale: Produce Freshness Detection

Every day, thousands of perfectly edible fruits and vegetables are disposed of because they fail to meet cosmetic standards. Picture-perfect produce is an unrealistic standard that leads to large amounts of food waste. Some supermarkets are working to reduce this waste by donating surplus food and selling “ugly” fruits and vegetables at a discount.

Milan Ferus-Comelo’s SmartScale aims to introduce an objective standard for “evaluating and selling fresh produce.” Since appearance doesn’t tell the full story, it uses radar imaging and AI to “assess the true internal and external condition of fruits and vegetables.”

It works as part of a retail checkout experience where a customer places a fruit item on the SmartScale, and the device calculates a freshness score using the built-in neural network. The SmartScale applies a proportional discount based on the score and prints a price label for the customer. It reduces waste and enables customers to make sustainable decisions.

It currently works with bananas, but can be extended to more fruits and vegetables.

THEIA: Tracking Heuristic Evaluation Intelligence Analyzer

Situational awareness is a crucial part of home security and automation. Bryan Staley and Brayden’s project uses the PSOC 6’s radar sensor and an edge AI model to analyze doorway traffic and classify passing objects.

The project comprises three separate applications for training, inference, and processing: THEIA-REC records and downsamples training data, THEIA-INF runs model inference, and THEIA-CLIENT processes the results and displays the information to the user.

Due to time constraints, the model has two main classes: person and ball. It can detect a person and a ball moving in and out of the radar sensor’s field of view.

AI Blender Speed Detector

Guillermo Perez Guillen's project detects the current status of a blender using the PSOC 6’s onboard accelerometer and gives a corresponding voice or LED alert. to detect when the blender is in operation. The idea can be applied to any device that operates with DC motors in any setting.

The board had to be placed in different orientations to capture at least two axes of the accelerometer. The project supports LEDs and voice alerts via the EDU DFR0699 voice recorder module.

Illegal Logging Detector

Illegal logging is a major issue in most developing countries. The Congo Basin, for example, loses 1 to 5% of its forest cover every year due to illegal logging, and a large proportion of wood consumed in Mexico is of illegal origin, according to a 2025 study.

Alejandro Sanchez’s device detects chainsaw sounds and human voices in protected areas, signaling that illegal logging may be occurring. The PSOC 6’s onboard microphone obtains an audio signal, which the model classifies into one of three classes: chainsawI, people, or forest (base state).

IntelliFan

There is nothing quite like a cool breeze blowing in the heat of summer. Most fans blow in a single direction and need manual adjustment. Wouldn’t it be cool if your fan followed you around the room rather than the way around? That’s why Ritik and Zilu of De Anza College built IntelliFan, a fan that tracks people automatically and responds to hand gestures.

The project uses DEEPCRAFT's pretrained gesture model and PSOC 6’s radar sensor to recognize five distinct swipe and push gestures. The web interface displays a live camera view and has controls for speed, tracking, and manual positioning. The team plans to upgrade the current product with voice commands, a battery for portable use, and a tower design for better airflow.

Sarah Hemmer, product manager at Infineon and one of the contest's judges, commented on the diversity of the winning projects, "I was amazed by the creativity of the contestants and the variety of use cases that were tackled as part of the challenge."

DEEPCRAFT's edge AI software, ModusToolbox, and the PSOC 6 Evaluation Kit were crucial in bringing these "edgy" solutions to life.