How to start

The visual editor of the AI-Builder design studio allows building AI solutions, also called pipelines, in a quick and intuitive way. The studio provides the necessary tools for AI developers: a collection of building blocks for your solutions, a canvas to build on and visual guidance on how to connect the blocks. The AI-Builder design studio is a part of the open source AI-Builder platform, running on a customized Acumos instance, a platform and open source framework. The platform supports the development of AI solutions as portable, containerized micro-services. Users are free to choose any modelling language, toolkit, run-time infrastructure or cloud service. The studio is open to anyone, you just have to create a user account and sign in.

Collaborate

The design studio invites collaboration: You can create groups for working together on building AI pipelines; you can share individual building blocks as well as complete solutions with fellow platform users. For each shared resource, you are able to specify the terms of use in a commercial or non-commercial license profile. In this way, the AIoD platform facilitates knowledge transfer. Businesses can share use cases or data and get support from European AI developers, AI experts can test and train their models with real world data, and researchers get inspiration from application scenarios.

Share your solutions

We invite users to create and upload their own building blocks to the AI resources catalogue and use these in the design studio as well. Anyone who has registered and created a user account for the AI-Builder platform can upload resources, which are checked for quality by our experts and then published in the resources catalogue.

Once your solution is ready, we recommend on-boarding it as a containerized micro-service, for example as docker container. It is not even necessary to upload the container; instead, you can provide a link referencing the storage facility, such as a docker registry. In that way, users can easily access the solution and deploy it on any execution environment.

Build hybrid AI-solutions

The AI-Builder platform facilitates the creation of hybrid and modular AI solutions. When building your pipeline, you can choose from the re-usable, existing building blocks, for example AI models for object recognition, classification or segmentation. Hybrid solutions combine different AI technologies, for example classic Machine Learning with symbolic AI, reasoning or constraint programming. Combining two or more AI methods is beneficial in some application scenarios, especially when data sets are incomplete. Moreover, this approach facilitates the development of human-centred AI, which meets the highest standards with respect to reliability, explainability and trustworthiness.

Example 1: video object recognition

The main building blocks of the solution are two models: video segmentation and video object recognition. A video file broker manages the data input. This basic pipeline is further enhanced by auxiliary infrastructure nodes depicted on a purple background: a persistent volume provider on the left, a model initializer and a tensor board connector, a tool for analyzing a model.

Example 2: Speech-to-text pipeline

The second example shows a pipeline that converts audio files to text format. The input data can be any spoken content, such as an interview or a discussion. The audio file broker on the left manages the audio data input. The audio segmentation model then cuts the data into short segments and eliminates pauses. The audio-to-text model is the core element of this pipeline: In our example, the model converts German language speech content into text. An auxiliary model adds punctuation such as commas, full stops or exclamation marks. Finally, the audio dialog creator assembles all segments in the right order and provides the output in text format.

Two other models are optional additions: The audio speaker recognition identifies individual voices and thus attributes each audio segment to a particular speaker. Thus, the text output has the format of an interview transcript or a screenplay where individual roles are indicated. Finally, the audio topic extraction model automatically extracts keywords that are representative of the content. In this way, the resulting text output is ready for tagging and automatically generating abstracts.

The speech-to-text pipeline is an example of the modular approach to building AI solutions, relying on re-usable building blocks. Developers can easily replace the audio speech-to-text model for the German language with any other such model for a different language, for example French. The audio punctuation model would then also have to be adjusted. All other models and elements of the pipeline can remain unchanged.