AI-POWERED DASHBOARD BUILDER
Beyond the dashboard: Designing AI that Guides, Fixes, and Builds
AI-powered dashboard builder that streamlines data exploration for users with varying technical skills. By embedding AI as a collaborative guide, not just a code generator. It transformed a manual, code-heavy process into a more intuitive, supportive experience.
role
Solo Product Designer
skills
Product design
User research & testing
Prompt testing
Roadmap planning
Interactive prototyping
TIMELINE
Design: April 2024 - July 2024
Launch: December 2024
OUTCOME
28%
Task success rate increased
40%
Task completion time reduced
97%
Active users adopted the AI feature
31%
Dashboard creation increased
COMPANY Background
Ludis Analytics is a SaaS platform for data preparation and deploying apps to generate insights. It supports national sports teams globally, including in the Olympics. While technical users could navigate Ludis comfortably, semi- and non-technical users often struggled with manual workflows and technical barriers.
Problems
Manual and error-prone processes
Despite the rapid evolution of generative AI, many of our users still rely on manual, code-heavy methods for data exploration and dashboard building.
Business pressure
At the same time, investor expectations were rising.
We were being asked: “How is your platform leveraging AI?”
Our stakeholders were eager for innovation—not only to improve product value but to broaden our user base beyond just highly technical users.
challenge
Ambiguity
We saw an opportunity to modernize the workflows—but the right path forward wasn’t obvious. There were many possibilities, and we needed to understand where AI could truly make a difference. With limited resources, we can’t chase every direction. The challenge for me is to identify the right AI features and scope out the MVP.
user segment
Who are the target audience?
To progress in clarifying our scope, we must first determine the audience of our AI product. Our team took a step back to revisit our users and identified semi-technical users can become more independent and build more complex analyses.
USER INTERVIEW INSIGHTS
What are our users struggling with?
We conducted 5 interviews on semi-technical and technical user and noticed 3 themes
Trial-and-error in visualization
I usually just try random charts until something looks close to what I need
Rigid, static dashboards
Once the dashboard is published, it’s hard to adjust anything on the fly
Dependency on technical teams
We met three times just to get the chart "right"
User journey map
Building reports as the main focus
We mapped out the end-to-end data analysis journey and decided to focus on the “build report” phase where we saw the most potential to deliver value. Not only was it one of the most frustrating steps for both technical and semi-technical users, but it also aligned with our platform’s current core feature - an inline code editor built for dashboard deployment.
experiment
Rapid prototyping to validate direction
To get the full story of the dashboard building experience and spot opportunities for AI, I personally build a dashboard using only my limited coding skills and AI support. This hands-on process revealed key friction points and helped us identify ways to offer more meaningful AI support.
Lack of a clear starting point
Without contextual guidance or a clear starting point, I had to rephrase my prompts multiple times just to find direction. Often, the more I tweaked the prompt, the messier the responses became.
Absence of visual previews
Instead of showing a visual preview, the system returned static, text-based output. A visual starting point could have sparked inspiration and built confidence faster.
Complex technical instructions
The response I got back often turned into walls of texts: package installations, environment setup, and data linkage steps. This felt overwhelming and created unnecessary friction—especially when those capabilities already exist in our platform.
Lengthy errors
When transferring code into our platform, I was quickly met with confusing error logs and cryptic messages. Without clear direction, it was difficult to know what went wrong or how to fix it.
Prioritize features
Strategic prioritization and define MVP
After the experiment, I listed out all the possible features that can smoothen the experience. But as a startup, we don't have the resource to chase after every idea. So I worked closely with developers, we evaluated each idea based on engineering lift, user value, and targeting user counts to build out the MVP for this product.
Design explorations and decisions
How automated should AI be?
One of the most debated questions in our team was: “How much automation is too much?” In early prototypes, I explored an option to place a “Run” button directly in the chat interface, so users could execute AI-generated code instantly. However, this introduced a few challenges:
It encouraged users to stay in the chat interface and bypass our core product UI
It reinforced a dependency on AI instead of gradually building familiarity with our platform’s features.
Instead, we chose a lighter-touch approach: AI responses would guide users to locate and use the platform’s native “Run” button. This subtle nudge supported our larger goal: building user confidence and bridging the gap between non-technical and technical workflows.
Initial iteration
Run code directly trough chat interface
FINAL iteration
Prompt users to use platform's native "run" button
Design explorations and decisions
How to build trust?
To build trust in AI-generated insights, we added an inline explainability interaction for each suggestion. When the system recommended a chart or query, users could hover over a small “Why this?” tooltip to reveal the AI’s reasoning. It helped users understand the logic behind the AI, reducing feelings of randomness.
Design explorations and decisions
How to protect data privacy?
Another design exploration involved making dataset linking easier. Initially, I tested UI solutions that would auto-detect datasets and pass them to the AI model behind the scenes. But this raised data privacy concerns. We didn’t want our AI model to access live user datasets directly, especially for enterprise accounts.
To avoid privacy concerns, we prompted users to manually paste dataset URLs into the chat when starting an exploration. While it introduced a small extra step, it was the right trade-off between usability and platform security.
Design iterations
Improving the experience
There were a few usability issues were noticed during user testing, we later iterated quickly on the chat interactions to improve the viewing experience.
Iteration on improving focus with tabbed view
In the original layout, code appeared above the preview within a single output area. This forced users to scroll frequently and broke their focus. I redesigned on the layout by combining the code output and the preview panel into a single section using a tabbed layout, allowing users to easily switch between building and viewing without jumping between separate context.
Initial iteration
Separated sections for code and preview
FINAL iteration
Add a tabbed layout to combine code and preview
Iteration on scrolling interaction for smooth reading
Users had to scroll up frequently as new content loaded below their view. I observed this in a few usability sessions and flagged it to the dev team. We quickly updated the chat layout so that the newest message would anchor at the top, making scrolling behavior feel smoother and more intuitive.
Initial iteration
Constant scroll up motions to read a long message
FINAL iteration
Anchor the message at the top to allow smooth reading
FEATURE HIGHLIGHT
Ludy - AI Dashboard Builder
Users often didn’t know how to begin exploring their data. They struggled with prompt phrasing, got overwhelmed by messy outputs, and lacked confidence in the code they received. We designed a multi-layered experience that guides users through the dashboard creation process with clarity and confidence:
Quick insight
Instead of starting from scratch, users can ask natural language questions and get instant answers like key drivers or standout values. This speeds up early exploration and helps users quickly find direction in complex datasets.
Insight overview
The assistant automatically points out trends, outliers, and correlations which guides users toward meaningful insights they might have missed. It acts like a built-in analyst that adds context to their exploration.
Code preview
To build trust in the AI-generated code, we designed a live preview that shows what the dashboard will look like before users commit. This lowers the barrier to entry and helps users feel more confident and in control.
FEATURE HIGHLIGHT
Log Summarizer
Dense logs overwhelmed users and slowed down their momentum. We turned this pain point into a moment of clarity by designing a log summarizer that translates complex logs into plain language with recommended actions. Instead of scanning lines of code, users now instantly understand what happened and what to do next which empowering them to stay in flow without calling in technical help.
FEATURE HIGHLIGHT
Debug Assistant
Getting stuck on bugs used to stop progress. Our Debug Assistant turns that moment into an opportunity. When errors pop up, the assistant suggests fixes, rewrites broken code, or offers next-step guidance—all within context. It’s like having a technical partner built into the workflow, giving users the confidence to solve problems independently and keep building.
IMPACT
Within 3 months of rolling out these updates:
28%
Task success rate increased
Ludy AI reduced confusion and guided users through complex steps, helping more users complete tasks successfully on their first try.
40%
Task completion time reduced
AI-generated insights and code previews helped users skip trial-and-error which speeds up the whole process.
97%
Active users adopted the AI feature
The seamless integration made the platform more intuitive, driving high adoption across technical and non-technical users.
31%
Dashboard creation increased
With clearer direction, easier debugging, and faster iteration, users were empowered to build more dashboards more often.
REFLECTION
Designing this AI feature required constant trade-offs between ease of use, system limitations, and building long-term user trust. By simulating user behavior, gathering feedback from beta testing, and partnering closely with engineers, I made product decisions that solved real pain points while supporting our broader business goals. This project sharpened my ability to turn complex workflows into meaningful, adoptable solutions.
