ABI - BID - Revit, Experience Design
Revit Copilot (AI Assistant Exploration)
Autodesk Assistant
My Role
Lead Designer - Exploration & Conceptualization
Goal
To explore, conceptualize, and define the user experience for a generative AI assistant ("Revit Copilot") within Autodesk Revit, establishing a vision that is both aspirational and grounded in user value.
Designing the Future
As the Lead Designer for the exploration phase of "Autodesk Assistant For Revit" (The project was called “Revit Copilot” at that time), I was tasked with a blank slate: How might we integrate a generative AI assistant directly into the complex workflow of a Revit user?
Background
Requirements From PM
- Beyond a Chatbox: Move beyond the common chatbot paradigm to create an experience that is truly contextual and integrated into the Revit canvas.
- Establish Trust: Design an interaction model that feels reliable and trustworthy for mission-critical architectural work.
- Demonstrate Value: Create concepts that solve real user problems, proving the assistant was a valuable collaborator, not just a feature.
- Combine Notification System Into the Chatbox: Notification Panel is a project has been hold in Revit for a long while. PM would like me to explore the possibility of combining the Notification System into the AI Tool.
- Pave the Way: as the project will be transferred to other designers and scrum teams, PM asked me to pave the way for the upcoming design work. This means identify as many risks and great ideas as possible.
As you can see from the below image, this is the current Revit Interface (though a bit exaggerated). Users can always see nested dialogs in Revit. So, where should the AI Chatbox fit into?
01
Exploratory Design & Concept Generation
Leading the Conceptual Phase
I led the translation of high-level strategy and research insights into tangible experience concepts. My work focused on exploring the "how" and "where" of the AI interaction.
Key Design Explorations I Drove:
Use Case Visualization: Creating wireframes and flow diagrams to bring the highest-value use cases (like automated documentation or troubleshooting) to life. Below is a design for connecting error handlings with AI Assistant.
Integration Patterns: Exploring how and where the AI lives in the UI—from a persistent panel to contextual, floating elements.
Cohesion with Platform: Ensuring our concepts for "Revit Copilot" could align with the broader Autodesk Assistant platform vision and design system. Below is a screenshot of the research I done on the Assistant Interface in other Autodesk Products.
02
Synthesizing a Cohesive Vision
Defining the Experience Principles
Through rapid iteration and collaboration with research and product manager, I synthesized our explorations into a set of core experience principles that would guide the project.
The Vision We Defined:
The AI Assistant should be Contextual, Predictable, and Empoweringly Simple:
- Contextual: It understands the user's selection, view, and task, offering relevant suggestions without being asked.
- Predictable: Users should have a clear mental model of what the AI can and cannot do, building trust through transparency.
- Empoweringly Simple: It reduces complexity, making powerful functionality accessible through simple language or clicks.
03
Outcome & Transition
Blueprint for the Future
The exploration phase concluded with a clear, user-validated design principles and a robust set of experience concepts other designers can adopt. This work was instrumental in making the strategic decision to invest significantly in the project, moving it from exploration to execution.
Project Evolution
Following the exploration phase I led, the project was formally transitioned to a dedicated, cross-functional team (under the official name "Autodesk Assistant for Revit") for detailed UX/UI design, engineering implementation, and further validation.
Key Learnings
- Design-Led Exploration: A designer can and should lead the initial phase of a complex, ambiguous project, using visual concepts to make the future tangible and align stakeholders.
- Advocate for the Experience: My role was to champion the user's needs and the integrity of the experience, ensuring technical decisions were made with the end-user workflow in mind.
- Laying the Groundwork: The most impactful design work is often strategic and foundational, creating the blueprint that guides all subsequent execution.
04
Beyond the Project: Prototyping the Future of Help
Extending the Vision: Personal AI Exploration
My work on Revit Copilot sparked a deep curiosity about the practical mechanics of AI, particularly RAG (Retrieval-Augmented Generation). To move from theory to a more tangible understanding, I've been actively exploring how to build and implement these systems myself.
My Current Learning Project: A Conversational Help Agent
I am currently using Cursor (an AI-powered code editor) and n8n (a workflow automation tool) to experiment with building a prototype AI agent. The goal is to create a system that can provide better answers from the vast Revit Help documentation.
How It Works
- Knowledge Base: I processed the entire Revit Help documentation into a vector database (using Cursor to crawler the entire help content as our Content designer said there was no text file can be shared with me lol), allowing the AI to understand the semantic meaning of questions, not just keywords.
- User Query: A user asks a natural language question (e.g., "How do I create a multi-story stair?").
- Intelligent Retrieval: The AI agent finds the most relevant sections from the help docs based on the question's intent.
- Contextual Answer: The LLM synthesizes the retrieved documentation into a clear, direct, and cited answer for the user.
The Objective
This hands-on exploration is not about building a shipped product, but about deepening my practical knowledge. By getting into the details of how AI assistance is constructed, I'm better equipped to design feasible, impactful, and user-centered AI experiences for the future, directly informed by the foundational work I did on Revit Copilot.
But theoretically speaking, it my practice works, the workflow can be reused in any projects as long as it has a thorough Help Documents.