Porsche
Dream Cube
Engagement by machine learning
Role
Tech concept und interaction design, code, model training
01
Task
Your dreams are our content
At the end of 2022, Porsche opened the first Brand Store in the world in downtown Stuttgart. Not just as a showroom, but as a lively center for the real dreams of real people. Before the opening, we wanted to show people the uniqueness, beauty, and value of their dreams and convince them to turn their dreams into reality. Our solution: a spectacular out-of-home installation.
02
Installation
Dreamer welcome.
In the heart of Stuttgart, we provided a 3x3x3 meter LED cube as a giant canvas for their dreams. Visitors entered their dream data directly on their smartphones. The low-threshold interaction captivated many passersby. A specially developed content management system transformed their dreams into breathtaking animations before their eyes.
03
Interaction
May we have your dreams?
Special radar sensors enabled live interaction with the content, breaking the boundaries of traditional interaction. Participants could download their dream videos and share them on social media, increasing digital reach and engagement. From the first contact to the event at the Brand Store, we developed a cohesive story about their dreams.
04
Emerging Tech
Transformer (before they were cool)
The dream visualizations were driven by individual user data, which influenced the color, dynamics, and thus the uniqueness of the animations. The dreams were analyzed with a BERT classifier that generated metrics on the following aspects: Boldness – Big and ambitious dreams led to higher values and energetic colors. To distinguish bold dreams from small private wishes, we trained the model with 30 standard works of futurology. Engagement – Dreams formulated in a compelling, encouraging way received higher engagement values, accelerating the visualization. The training material consisted of a large corpus of texts with an activating tone.
05
NLP
BERT in da house.
The project utilized Google's first practical language model BERT, which uses the transformer architecture for training and classification. The implementation was a particular challenge as the originally English-language model had to be fine-tuned with large amounts of German-language texts. At a time when transformer technology was just beginning to take the tech world by storm (The "T" in GPT stands for Transformer.)
06
Visualize
Controlling unreal engine and speed
Data and interaction formed the basis for live rendering with the latest Unreal Engine, which orchestrated the colors, shapes, and dynamics of the animation with tens of thousands of particles. These complex calculations remained invisible to the dreamers, who delighted in the unique and aesthetic representation of their personal dream content.