At Netgen, we constantly experiment with new AI tools to see what impact and challenges they bring to our employees and their workflows within a digital agency and what this means for our clients.
To give you an insight, in this blog post, we are preparing some AI use cases we have been working on recently to share the concepts and thoughts behind them as well as the results and insights on how we have improved and automated workflows with AI.
Before we start, I would like to point out that the AI tools we use are changing and developing fast and give a general appraisal of their expected evolution:
The first phase of AI tools of almost pure, playful exploration of AI tools and their capabilities for mostly entertaining fun projects and applications has already moved on to the second phase of controlling outputs within individual AI tools, paving the way directly to the third phase of automating standards of linking multiple AI tools to speed up workflows. In the final step, this will lead to phase four, in which autonomous systems will work without direct human input in the future.
Even though autonomous AI agents are not yet available, we want to give insight into how Netgen applied the first three phases by combining controlled outcomes with automated workflows to develop creative thoughts, ideas, and concepts and bring some concrete project examples to life:
1. Automated and networked workflows with AI agents
The systems' capabilities and connectivity are increasing, which means that more and more work processes can be connected and automated with the help of AI. The improvements are bringing us step by step closer to AGI (artificial general intelligence). The upcoming autonomous AI agents will solve tasks autonomously – independent from human interactions – in the future.
«AI is already automating the content creation, and will do so autonomously in the future.»
Amar Delić, AI geek & Senior IT project manager, Netgen Switzerland
Amar Delić, our AI geek and Senior IT manager is already assigning subtasks to several agents and models and linking them together to (automatically) solve the specified main task.
2. "AI minutes", the AI-generated AI podcast
Idea
Our AI explorer Dennis Oswald and AI geek Amar Delic put their heads together. They conceptualized, designed, and developed, using AI tools wherever possible – from the concept, the naming, the visual elements, the text modules, and, of course, the performing voice.
The result is AI minutes - one of the first AI-generated AI podcasts in Switzerland, compiled and presented by an AI presenter – produced by Netgen Switzerland AG.
Concept
For the exchange and development of ideas, we decided to set up an internal, customized GPT (OpenAI) and "brief" it according to its role as a sparring partner in creating an audio podcast.
Using this "AI Podcaster" GPT, we created via "Prompting":
- The conception of the podcast in general and the corresponding episode structure
- The development of the text modules, e.g., intro, outro, boilerplate, and drafts for social media posts.
- Naming, which was driven forward and worked out in dialogue with ChatGPT/GPT
- Prompt engineering for input into other AI tools, e.g., generating visual elements or sounds. (text-to-image; text-to-sound)
Visuality
The first images for the later key visual were generated directly in dialogue with the GPT. Based on their visual style and prompt content, work continued in Midjourney. The biggest challenge apart from creating the visuals is the sheer number of variants, which can quickly get out of hand due to small changes in the prompt.
Voice & Sound
The voice, the central element of the audio podcast, was created using the Text-2-Speech tool from Elevenlabs. Here and there, she - Lily - breaks out into artifacts of German, especially with longer passages of text or English terms. The new audiobook solution "Projects" (beta) from Elevenlabs promises to remedy this in the future.
The text-to-sound tool waveformer with the MusicGen model on Replicate was used for the background music.
Implementation & API
The most exciting step in creating the podcast is the automation and coordination between the various AI tools. The choice fell on Python and Streamlit for the graphical user interface (GUI).
The aim is to automate as many steps as possible while allowing human intervention at the most critical nodes. At the moment, the content of an episode is (still manually) collected and imported via GoogleSheet.
The language of the source no longer plays a role in creating the episode transcript. The audio file with the selected voice is generated automatically via the API connection, as is the background music in the next step.
Our tool can be configured via the GUI and can control the various agents that work together to create the podcast. The internal workflow is as follows:
- Configuration of parameters and services (APIs)
- Scraping of the individual news texts and preparation
- Generation of the podcast episode text with the OpenAI GPT4 Turbo model via the API
- Speech synthesis of the podcast via the ElevenLabs API with the selected voice
- Mixing of voice and background music with generated podcast and level adjustments
Anyone interested in further technical details is welcome to contact our AI geek, Amar Delić .
Summary
The AI minutes audio podcast is a relatively small-scale test project for Netgen Switzerland. However, it shows the enormous potential of controlled and automated AI processes in the future.
- The costs of artificially generated content will fall, and the opportunities that arise from this will transform the overlapping areas.
- Automated, comprehensive searches and summaries - whether text, sound, image, or video in any length and quality.
- Language as input or output is no longer an obstacle and can be customized and scaled.
- Other AI tools can be connected, e.g., for post-processing or distribution.
- The process is transferable to other disciplines, media, and workflows.
At the moment, it (still) requires human intervention here and there, and a human must check the results produced, but it is also clear that the rapid development of these tools means that more and more tasks will be automated. So the question remains that we, as a society and as individuals, have to answer for ourselves: If there are no longer any limits to what is technically feasible, where are the limits to what is socially and morally acceptable?
3. AI plugin for CMS
All our websites are CMS-based to give the creators the flexibility to change, translate, and update the page elements and content independently of us.
Especially for our Swiss clients, multilingual sites are a must because of the four national languages (German, French, Italian, and Rhaeto-Romanic). In addition, the content of an article, for example, often comes in pieces from several sources or even from authors in different languages.
That's why we have developed a plugin for the Ibexa CMS that enables seamless integration of AI into content creation and maintenance.
The Netgen plugin allows jo to:
- automatically generate text-to-content
- allows to fetch other fields as a source for content generation
- automatically generate images based on content inputs
- fix spelling, extend, shorten, rephrase, or translate given texts
- or translate automatically whole content sections
Furthermore,
- it can be connected to various online AI engines (such as ChatGPT, Llama, Claude, etc.)
- it automatically highlights which fields are generated with AI to give control to the creator
- it is highly configurable, and the user can easily change AI prompts for the content creation.
4. Creating Landing Pages with AI
Automating the creation of landing pages is in high demand from our customers. The potential to do this with the help of AI is enormous.
The core task is to set up the AI to (only) use and create existing and updated content from the specific platforms. This can be achieved by using Retrieval Augmented Generation (RAG) to retrieve facts from an external (corporate) knowledge base to aim for the most accurate outcomes of the connected large language models (LLMs).
Idea & concept
This prototype showcase demonstrates the power of combining a custom-trained GPT, web scraping, and modular web design to create personalized, content-rich websites efficiently.
First, we set up a GPT model and trained it for landing page creation. We used web scraping to extract data from a corporate platform and fed this information into the GPT using the Retrieval-Augmented Generation (RAG) technique.
The project mainly focused on leveraging the GPT's capabilities to generate custom, detailed landing page structures with rewritten and curated content. By providing the GPT with the necessary context and data, we enabled it to create optimized, high-quality content tailored to the website's specific needs.
The next step in the project is to design a web UI based on modular layout blocks that align with the GPT-generated landing page structure. This modular approach allows for a flexible and visually appealing design seamlessly incorporating generated content.
Once the UI is set up, the final step is to populate it with the GPT-generated content, ensuring that each landing page section is filled with relevant, engaging, and optimized information.
This prototype showcases the potential for automating website creation by integrating AI-driven content generation and modular web design. By leveraging the power of GPT and web scraping, businesses can create high-quality, personalized websites with minimal manual effort, revolutionizing how websites are built.
5. Summary
We at Netgen believe in the added value of integrating artificial intelligence into several steps of our workflows to improve the quality and quantity of our project work and manage them.
This does not mean we want to substitute human labor with AI; it means empowering our teams and clients to adapt quickly to the new available skills and tools and the speed of their developments to work smarter and focus on the more exciting parts of the workflows.
We recommend checking out our landing page, Unlocking the Power of AI for Your Digital Transformation, to find out more about our approaches to harness the potential of AI for your digital transformation journey, such as:
- The Use of AI in a General Digital Strategy
- AI in the UX & Design Process
- AI in Development (e.g., Tune AI Models)
- AI in Marketing Automation and Growth (e.g., AI Predictive Analysis)
- AI Training and AI Consultancy
- AI-Powered Content Generation