Rule-based chatbots, rebuilt into a controllable AI agent platform.
Userbot started a strategic redesign to bring LLM capabilities into the core of its product, changing how conversational automation is built and managed. As a freelance Product Designer, I supported this transition end to end, rethinking the platform's architecture and user experience to support more intelligent and flexible interactions while keeping the trust, reliability, and clarity users needed. My role was to turn this technological leap into a coherent, scalable product experience.
Making abstract agent logic feel concrete.
Adding LLMs to an existing platform meant every design choice had to balance new AI intelligence with the trust users needed.
Agent concepts like routing, retrieval, and variable extraction had to be translated into clear, consistent interaction patterns that non-technical users could trust.
Document-specific configurations and specialized tools risked fragmenting the interface; the system needed to stay consistent and flexible long term.
Existing manual bots and new AI agents had to live side by side seamlessly in the same platform, without breaking established workflows.
Every surface, from the Workspace to the Flow Builder, had to remain intuitive as capabilities and projects multiplied.
The decisions that did most of the work.
Centralize knowledge, keep setup consistent.
I introduced a centralized Knowledge Base: the single place where all company knowledge lives. Users can upload company documents in any format, add links to web pages, or write Q&A pairs. The AI agent uses all of this content to understand context and answer customer questions. I designed a different setup for each content type, all within one simple, scalable interface.
Turn agent logic into modular blocks.
I redesigned the Flow Builder around modular agent blocks such as routing, retrieval, Q&A, and variable extraction, and introduced a scalable "toolbox" layer of agent tools to extend specialized capabilities within workflows.
Redesign the front door.
I redesigned the Workspace page where users select and manage their AI agent projects, making it easy to switch, access, and manage projects at scale. Video tutorials and ready-to-use templates make it simple to get up and running, and help users explore the platform's full potential.
Rebuild the Flow Builder's core commands.
I repositioned and redesigned key actions like navigation, preview, the search bar, and workflow management, so that existing manual bots and new AI agents could work together smoothly in one simple, consistent system.
What this project taught me.
A few principles carried through the project: reframing the problem, designing for coexistence, and building for what's next.
The real project was not adding AI features but redefining the platform's mental model: from scripting conversations to configuring goals and behaviors. Identifying this shift early kept every flow decision consistent.
Manual bots could not disappear overnight. Designing for coexistence, rather than replacement, made the transition credible for existing customers and safer for the business.
Modularity was the answer to abstraction: whenever agent logic felt too abstract to design for, breaking it into blocks and tools gave both users and the team something concrete to reason about.
The consistent setup patterns and toolbox layer were designed to absorb capabilities that did not exist yet, which is what an LLM-powered product ultimately demands.