Edge Data

Customize your edge AI solutions with low-code workflows

Today, AI can be found everywhere, not just in the cloud. AI is being increasingly used at the edge.

Today, AI can be found everywhere, not just in the cloud. AI is being increasingly used at the edge.

Edge AI solution providers need to deliver customized AI solutions for their end customers. This often requires significant development time and effort and ultimately, slows down solution adoption. This is particularly challenging in the industrial sector, where there are many data sources, data protocols and unique application needs. These variations require AI solutions to be customized.

Customize your edge AI solution

Where is customization needed for edge AI? We've gathered for you the most common places for customization in your edge AI data pipeline below.

  1. Meet varying customer needs when acquiring data

    Data acquisition can come from a variety of sources such as sensors, machines, historians, API endpoints, MQTT endpoints, to name a few. The format of the data produced by these data sources are also varied, including JSON, XML, CSV and other file formats. These variations make it nearly impossible to build a single solution to satisfy all customers’ needs.

  2. Stay above the noise by cleansing data

    Data cleansing is necessary because edge data sources like sensors and measurements can be noisy and sometimes unreliable. Performing data cleansing helps prepare the data based on the property of the data and the application's desired outcome.

  3. Providing consistency for data massaging

    Edge data is usually time-series data, but more often than not input data sources produce data at different time intervals. In these cases, interpolation or filling is needed to ensure data consistency.

    Free Resource: Dive deeper into your time-series edge data pipeline with InfluxDB

  4. Evolve quickly during data transformation

    Disparate data source formats need to be converged to a unified schema, and metadata needs to be added to identify each data source. Since data formats can change and new sources can be added, existing edge applications need to be agile to stay on top of these changes.

  5. Better machine learning with feature engineering

    One way to improve machine learning is through feature engineering. This is the process of using domain knowledge to extract features from raw data. Techniques such as averaging, scaling, normalizing, clustering, and one-hot assist the machine learning process, but also require customization.

  6. Get more from AI inferences in post processing

    An AI inference output can include a lot of information, and specific properties may need to be extracted depending on the application. For example, a machine vision inference output may include every object the AI model detects: cars, persons, windows, doors, motion, and etc.  One customer may want to know the number of cars, and another customer may want to count the number of people entering the scene, so different properties need to be extracted for each customer.

  7. Extend and expand with further integration

    AI inference output may need to be integrated with other sensing, control, or software applications. For example, if an AI model predicts a potential machine failure, one customer may want to receive an email, and another may want to receive a call. One customer may want to send the result to a PI Historian database via API, and another customer may want to send the result to their cloud service using MQTT. There are many possible application options for each customer.

The downside to customization

Due to the endless options at both the input and output side, AI solution providers spend significantly more time implementing these customizations than working on their AI model. This slows them down and distracts them from their core competency, which is AI.

The benefits of low-code edge AI solution

One way to tackle this challenge is with low-code, which is a programming paradigm that is graphical, modular, and can implement 12x faster than full-code development. For solutions that involve many protocols, endpoints and databases, low-code is perfect because it can handle complex and diverse data operations. Drag-and-drop workspaces make it easier to build, test, and deploy edge AI solutions. That way, AI solution providers can deliver solutions faster to their end customers while end customers also have control over their solution.

How does it work?

A great way to combine low-code workflows with edge AI processes is to implement each in its own container. The low-code container and AI container communicate with each other through an API. This ensures that the low-code component and AI component are independent from each other.

Once the architecture is set up, the AI container can be implemented by the AI solution provider’s core team, while the low-code workflow can be implemented by their application engineering team, the customer’s engineering team or a system integrator. Now, different teams can work on the same project in different ways without interfering with one another.

For edge AI applications, both the low-code container and the AI container need to be deployed to the edge. After initial deployment, low-code workflows can be continuously updated without having to update the low-code container, reducing bandwidth and delay constraints.

The perfect companion to edge AI

Low-code workflows make the edge AI application flexible and agile, making it easy and fast to customize edge AI applications without compromising on the data functions. That’s what makes low-code the perfect companion to edge AI.

Learn more about low-code edge AI solutions

Learn more about how Reality AI uses Prescient Designer to build their edge sensing solution for factory and process-industry asset monitoring. Read the full use case here.


Similar posts

Stay informed on IoT trends and insights

Discover the business and technology insights of IoT to accelerate the success of your digital transformation strategy. Sign up for our mailing list today.