This blog will not look like what you often see in tech reports — where someone says, “Oh, ideas just came to my mind, and bang, it just happens.” The reality was far more challenging, iterative, and often frustrating.
When I first rolled out this Apple-style frontend UI, it looked sleek, and I was genuinely excited.
I shared it with everyone, thinking, “This is really great!” But as time passed, I started to see things differently. It felt good at first, but… it was bland.
We realized that to create truly breathtaking design, we had to anti-bland. Like the masterpieces of the Renaissance, it wasn’t enough to merely replicate — we needed to define our own unique style.
We don’t want to be another AI that just meets high benchmarks. We want to bring aesthetics to the world, to people who have never encountered the concept of design. But when they look at it, they feel the sensation.
Think of it more like the opening scene of The White Lotus Season 2 — unforgettable, raw, and full of twists.
We also looked at the top-tier publications, fonts, color schemes, and layouts. Some of our favorite findings included:
- Italian italics paired with bold type
- Bento-style layouts that are modular and balanced
- Using background overlays on cover images
- Glassmorphism style
- Uses Tailwind CSS
These were design elements that had a lasting visual impact, but we quickly realized that recognizing beautiful design is only half the battle.
In a team setting, aligning everyone around a common vision of what looks good is a whole different challenge.
As for why we created an “Bland Design Evaluation”— the standards of beauty were just too fragmented and inconsistent.
The goal was simple: we needed to make sure everyone on the team knew what bland design looked like, so we could collectively avoid it.
I also went through hundreds of bad examples, highlighting what went wrong and discussing them with the team. It was an exhausting, repetitive process, but it eventually paid off.
After multiple rounds of refinement, we reached a shared understanding of what aesthetics meant for our project. It was like training the team through reinforcement learning.
How to define “Bland” :
- Color Overload:
- High Saturation + Large Color Blocks
- Cluttered & System Typography
- Layout Chaos
While we were making strides on the design front, we encountered another challenge: long context. Because deep research’s result is often very long.
If we started incorporating this into our design, it became clear that these large blocks of text could severely impact the visual appeal.
This was the point where I realized the “impossible triangle” in design: Aesthetics, Interaction Layout, and Faithfulness. It’s like trying to get someone to be meticulous, creative, and smart all at once — something always feels off. But over time, we found a balance through continuous iteration.
My daily routine was filled with testing and tweaking. Every small breakthrough felt like a victory, and I learned the importance of documentation.
By saving every experiment, I could revisit ideas and make improvements that eventually led to success.
Here are a few examples of our experiment prefixes:
- 05–01-interactive-coloroptimized-bento-faithfulness-v1
- 05–01-interactive-coloroptimized-bento-faithfulness-font_enhanced-v999999
Faithfulness (minimize hallucinations) was one of the most difficult aspects of our project. It was messy, it was dirty, but it was also essential. The issue of staying true to the original data, while keeping everything visually appealing, demanded a lot of our time and attention.
We realized we needed to bring in external help, so we integrated an AI Judge. This AI tool helped evaluate design decisions that would traditionally be handled by humans. It allowed us to speed up the process and maintain accuracy in the evaluation.
Aside from aesthetics, we also had to focus on the user experience — particularly for deep research users.
I personally don’t enjoy dense, overwhelming text, and I really dislike scrolling endlessly to find relevant information. So, we introduced a grid layout and navigation. It’s more than markdown thing.
With the new layout, content was broken up into multiple columns with clear, highlighted sections. There was also a sidebar navigation that allowed users to jump directly to any part of the report. This made the user experience smoother and more engaging.
We also realized that people are visual creatures. To improve the UX, we began using images, charts, and diagrams. These visual elements really helped make the content more digestible. As a result, my experiment names became longer and more descriptive:
- 05–01-interactive-coloroptimized-bento-faithfulness-font_enhanced-withcharts-mermaid-image-v1
- 05–01-interactive-coloroptimized-bento-faithfulness-font_enhanced-withcharts-mermaid-image-v2
These additions significantly enhanced the overall user experience. We now had not only clean design but also interactive features that improved engagement and accessibility.
But even then, something was still missing.
After a few rounds of user testing, we realized there was a sense of static-ness in our designs. People wanted more interactivity. So, we began experimenting with subtle JS animations. These animations weren’t meant to be flashy or distracting; instead, they served to guide the user’s experience.
For example, we added animations that helped highlight key transitions, making it easier for users to follow content as it appeared. These small touches of interactivity helped transform the page from a static report into a dynamic experience.
Once again, this was another round of trial and error. There were days where I spent hours testing different animations, only to find that none of them worked. But then there was that moment of discovery when everything finally came together, and the result was stunning.
Throughout this entire process, I couldn’t have done it alone. I am passionate about aesthetics and design architecture, but when it comes to faithfulness evaluation, I rely heavily on my teammates.
Faithfulness is a complicated concept to measure, and I’m glad that my colleagues helped with that aspect. This collaborative effort truly allowed us to tackle the most challenging aspects of our project.
In the end, we realized that teamwork is the real secret to success. The ability to collaborate, share ideas, and get different perspectives was what made our final design and functionality work so well. We each brought something different to the table, and together, we created something that was greater than the sum of its parts.
- Never Settle for Normal: Design is an evolving process. What may seem great at first will soon start to feel ordinary. Constantly push yourself and your team to define what exceptional design looks like and strive to surpass those standards.
- Align Your Team Around a Common Vision: Having great taste in design is only one part of the equation. Getting everyone on the same page and ensuring that your team shares a common aesthetic vision is crucial. The “Ugly Evaluation” document became our tool for alignment, ensuring we all agreed on what constituted bad design.
- Master the Impossible Triangle: Balancing aesthetics, interactivity, and faithfulness is a constant challenge. It’s the impossible triangle of design, and finding the sweet spot takes time, patience, and lots of testing. But when you get it right, the results are worth the effort.
- Iterate, Test, Repeat: The key to finding the right balance is iteration. Experiment, document your process, and refine continuously. Sometimes it’s the smallest adjustments that make the biggest impact.
- Teamwork Makes the Dream Work: No matter how good you are at design or architecture, it’s impossible to create something truly outstanding without help. Trust your team, collaborate openly, and combine your skills to solve the toughest challenges together.
In the end, our project wasn’t just about creating something aesthetically pleasing. It was about crafting an experience that was functional, visually engaging, and faithful to the underlying data. As we pushed the boundaries of what was possible, we learned that the journey to perfection is an iterative one. It’s about constantly questioning, refining, and evolving until everything finally falls into place.
As with all creative endeavors, the process was painful at times — filled with frustration, setbacks, and long hours. But when it all came together, the sense of accomplishment was unparalleled.
We hope that by sharing these lessons, others can avoid some of the pitfalls we encountered and create designs that are truly exceptional. Ultimately, our goal is to continue breaking the aesthetic limits, one iteration at a time.