Reimagining AI as a Thinking Partner
Bert is a computer vision copilot that sits in front of your whiteboard. He is friendly and resourceful, and named after Elmo's friend Bert from the Sesame Street. Not to be confused with BERT.
Role
Product Designer
Timeline
2026
Project Goal
Eliminate the "Blank Canvas" friction in whiteboarding.
The Tech
LLM → Computer Vision → Analysis
The use case
Whiteboarding is essential for high-level strategy, but what if you need an extra prespective. We wanted to answer this: How can we leverage existing AI models to bring the level of awareness you have in ChatGPT, but to the conference room.
How can we bridge the gap between abstract thought and structured visual systems in real-time?
Let's take the functionality of uploading an image to Chat-GPT for analysis into a real-time feedback loop.
An over simplification of this became something as simple as asking Bert "How many calories are in this can?" To which the model would correctly respond "0".
The existing models are good enough
Bert interprets natural language not as a paragraph, but as a hierarchy. I designed an interaction model where the user provides the "Seed" and the AI generates the "Structure."
- Intent Interpretation: Converting a prompt like "Map a supply chain" into a directed acyclic graph (DAG).
- Node-Based UI: Instead of static images, Bert generates manipulatable blocks, allowing for a hybrid of AI speed and human fine-tuning.
- Infinite Canvas Logic: A minimal interface that centers the content, reducing UI "chrome" to keep the focus on thinking.
Voice UI
There were several considerations when developing the TTS model for Bert. We wanted to keep it as low cost as possible, not just because of frugality, but because it felt a bit more authentic. We didn't want to trick the user into thinking that their AI was more than just that: AI!
Impact and Outcome
Bert is a case study in seeing how well you can utilize AI toosl to get a desired outcome you want. It has become the stepping stone of AI generated app within AI Vibe coder toolket (inception, if you will). More on this in person.