AI
Drupal as a prototyping tool to rapidly build a proof of concept
I've worked with Drupal in many ways over the past decade and I've seen incredible improvements and updates, especially with regard to the content editor and site builder user experience and user interface. This rung true during a recent project I set up. My goal was to create a Drupal 10 AI focused application for a consultant here at Horizontal. The idea was to build diverse artificial intelligence workflows using Drupal AI contrib modules.
The agile nature of open source combined with the collaborative nature of the Drupal community meant that Drupal was early in terms of providing extremely powerful AI functionality. This would help propel our project with accelerated development timelines, foster innovation, and ultimately lead to a more robust and versatile proof of concept in contrast to proprietary platforms that would need more time to ramp up and more developer hands-on.
Custom AI content workflows
The main focus of the project was content creation through custom AI workflows. I took some tutorials and read different guides to help me get up to speed with architectural decisions. One standout that really helped me get into the right mindset was Workflows of AI put together by Marcus Johansson, a maintainer of many of the AI modules we ended up using. Another was a presentation by Murray Woodman on creating augmentor personas.
In the end, I chose these Drupal contribution modules: OpenAI / ChatGPT Integration, AI Interpolator, ChatGPT Augmentor, and ElevenLabs Field. These tools let us create content from a prompt, create and use custom augmentor personas, chain prompts together, turn text into speech, turn text into images, and clean up and properly format content with the CKEditor AI plugin.
Consultant onboarding
After setting up the remote development sandbox, I gave our consultant, Jennifer Johnson (JJ), a tour of how Drupal works. We looked at content editing and site building. JJ had never used Drupal before, but she had worked with other content tools. As we talked, I saw that she understood the idea of "atomic content." This means she knew about content types, fields, structured content, and taxonomies.
Atomic content is content broken into small elements (known as atoms),
such as copy and imagery, which are dynamically assembled to create
targeted web pages, emails, or other marketing assets.
AI tools and workflows
Besides the basic content and site-building tasks, I showed JJ how to set up ChatGPT Augmentor personas with custom instructions. Augmentor would be used for complex tasks analyzing text and providing near real time feedback in a logical progression on a content edit page. Interpolator would be used for minor automated tasks such as generating a text summary or teaser for a piece of content. In addition, we would have DALL·E 3 image generation and text to speech audio generation.
Content types, fields, and workflows
Thereafter, JJ started building augmenter personas, content types, fields, chained interpolator workflows as well as very complex configuration settings. In the end, there would be 16 content types spanning a wide range of functionalities, including:
Content creation and optimization: covering areas like article writing, SEO enhancement, and readability improvement utilizing the Flesch Reading Ease score system.
Multimedia asset generation: facilitating the creation of text-to-audio content and AI-generated media entities.
Marketing and customer support: offering tools for creating marketing collateral, customer support knowledge bases, and social media analysis.
Specialized workflows: addressing specific needs such as job description generation, research assistance, and product information enrichment.
Technical improvements: Offering specialized tools for search optimization (Coveo programming)
With these AI-driven content types in place, we were able to streamline content workflows, improve content quality, and enhance user experiences across various segments. This was a significant step toward integrating AI capabilities into Drupal for an efficient and effective digital strategy for our rapid proof of concept prototype.
Progress
The wide range of content types JJ created ended up with a total of 90+ custom fields. She also built 5 custom taxonomies and mastered a wide range of site-building tasks, which is especially impressive considering she had never worked with Drupal before. I provided her with only the basic guidance, but she quickly grasped more advanced concepts, such as setting up entity references and creating a custom content moderation workflow. She also created 40 custom augmentor personas used in the various content types.
Augmentor persona example
Here is one example of a custom augmentor persona, Research Company or Topic.
Research Company or Topic
Role: system
Content: You are an outstanding researcher. You think step by step and
validate each of your results before you return them.
Role: user
Content: Create an in-depth research report about {input}. Format your
results as a marketing brief in proper HTML format. Include sections on
history, market strengths, current executives and anything else you
consider important for me to know. Validate your output before you
display it. Validate your output. Think step by step and
validate its accuracy.
{input}
Example output
Here is screen capture output from one of JJ's augmentors.
The Strategist's POV on AI, Drupal and Prototyping
"When we started this project, I didn’t think about prototyping as being one of the main benefits of learning Drupal, but it has been," JJ said. "Most enterprises aren’t really going to invest in technologies until they understand them, and understanding the technologies means licensing them, which can be expensive and time consuming.
"By using open-source software to rapidly prototype with AI, we can accelerate development because a lot of the business logic can be assembled in advance and with the right stakeholders at the table.
"We can be deliberate in our decisions without development resources waiting around to see what they can build. Most Generative AI applications that do content generation operate the same by combining author actions, triggers and prompts to generate AI outputs based on your content goals, creating a blog, or a product description, or an SEO analysis report – whatever it is.
"If we can get clients to understand those basic elements and how they all tie together to create compelling content, we can create fantastic user experiences with those outputs that drive real business value for our clients, and real benefits for our users."
Summary
Throughout this project, one goal for JJ was to present this project at DrupalCon which ended up being well received.
Since then, JJ has continued to build this out and give demos as well as start to have some internal users test it out.
In the process, I knew that Drupal had grown to such a high degree that one can now consider it to be an excellent tool for site builders and advanced content editors to rapidly prototype a proof of concept and that ended up being the case here. The UI and UX has evolved immensely to have made this possible.
With Drupal being open source and Drupal 10's ease of use for content editors and site builders, this really propelled our project forward in a very nimble and efficient manner. Drupal's flexibility allowed for rapid integration of AI-driven content types, while its user-friendly interface empowered JJ to adapt to these workflows. This combination of open-source adaptability and intuitive design accelerated development timelines, fostered innovation, and ultimately led to a more robust and versatile proof of concept.
On the horizon: Drupal Starshot
With the advent of Drupal Starshot (AKA "Drupal CMS"), Recipes, Project Browser, and Experience Builder, these initiatives will further enhance site builders and content editors being able to rapidly build highly functional proof of concept prototypes.
Resources
- Workflows of AI (Video tutorials)
- Uplifting content with AI: ChatGPT and NLP Cloud / Web Tools & Technologies / Murray Woodman (Excellent video presentation on creating augmentor personas)
- OpenAI / ChatGPT Integration -> now in AI (Artificial Intelligence)
- AI Interpolator -> now in AI (Artificial Intelligence)
- AI Interpolator OpenAI
- Augmentor AI
- ChatGPT Augmentor
- ElevenLabs Field
- OpenAI API
- ElevenLabs
- Drupal
- Drupal Starshot