I Rebuilt Intelligence Again Because It Still Didn’t Feel Intelligent
This is a slightly embarrassing problem to have.
The Intelligence app was called:
Intelligence.
And it did not feel intelligent.
The underlying model was actually one of the most interesting parts of the ecosystem.
Evidence strength.
Contradictions.
Retention.
Readiness.
Cross-subject patterns.
Interventions.
Plans.
Recommendations.
The interface had all the ingredients.
It still felt dull beside the products around it.
That bothered me.
Not because the app needed more glow.
Because visual hierarchy is part of meaning.
If the product is supposed to compress a huge amount of learner evidence into clarity, the interface should feel decisive.
It did not.
The first redesign mistake was treating theme like identity
I tried changing colours.
That helped less than I expected.
An orange accent could make the app orange.
It could not make the app feel like Intelligence.
That was the real problem.
Verbal had a learning personality.
Quant had a precision-performance personality.
Intelligence needed a different one.
Synthesis.
State.
Interpretation.
Direction.
The interface should feel like a system looking across everything and telling me what matters.
A palette alone cannot do that.
I kept comparing it to my writing site
This sounds weird because Mikhil Writes is not an education app.
But the website had something Intelligence was missing.
Richness.
Contrast.
Depth.
Colour used deliberately.
Sections that felt distinct without feeling disconnected.
The page had atmosphere.
Intelligence looked more like a software dashboard assembled from cards.
So I borrowed the principle rather than the design.
Dark first.
Stronger contrast.
More editorial hierarchy.
Colour with purpose.
A visual identity that feels expensive without turning into cyberpunk nonsense.
Purple became the identity
The final direction moved Intelligence toward a black-first interface with purple as its core identity.
Not Verbal purple.
Not Quant recoloured.
A deeper, more analytical use of it.
Then supporting colours could represent meaning.
Warm tones for attention.
Cooler tones for stability.
Different evidence states.
Different subject contexts.
The colour system became semantic instead of decorative.
That made the app feel more alive immediately.
The dashboard needed fewer equal boxes
This was probably the bigger change.
A page full of same-sized cards says:
Everything is equally important.
That is almost never true.
Intelligence should know what matters.
So the interface needed stronger hierarchy.
Current state.
Primary concern.
Best next action.
Evidence quality.
Then supporting analysis.
The layout itself should communicate priority before the learner reads every label.
That is what the older version lacked.
It showed information.
It did not direct attention.
Low-evidence states had to look intentional
A weak model state is difficult to design.
If the app does not know much yet, the interface can look empty.
So the temptation is to fill it.
Fake scores.
Default charts.
Generic recommendations.
No.
The redesigned Intelligence experience treats insufficient evidence as a proper state.
Not broken.
Not unfinished.
A known condition.
That means the empty state can explain what evidence is missing and what action would make the model more useful.
The lack of certainty becomes part of the interface.
Charts needed to explain themselves
A chart can be visually beautiful and cognitively useless.
So the new pass focused much more on:
Tooltips.
Readable labels.
Text alternatives.
Drill-downs.
Evidence counts.
Time windows.
Why a signal changed.
The chart should not be the conclusion.
It should be another way to inspect the evidence behind the conclusion.
That matches the whole product philosophy better.
The next action became more prominent
Intelligence is not supposed to be an analytics museum.
If the system notices something important and cannot help me act on it, the insight is incomplete.
So the UI gives stronger weight to the next action.
What should I do?
Why?
How confident is the system?
What would success look like?
Where does the action happen?
Usually the answer should send me back into Verbal or Quant.
That movement matters.
Intelligence understands.
The subject app teaches.
Then the outcome comes back.
Privacy needed a place in the product
As the learner model became richer, settings could not remain superficial.
The interface needed room for:
Data transparency.
Export.
Deletion.
Preference control.
Evidence status.
Model-related explanations.
An intelligent product becomes less trustworthy if the learner cannot understand the boundaries of the intelligence.
This is not a legal-footer problem.
It belongs in the product.
The app needed complete non-happy states
This was part of the wider three-client polish.
Loading.
Empty.
Stale.
Offline.
Error.
Insufficient evidence.
Locked.
Those states cannot be afterthoughts because Intelligence is especially vulnerable to stale or partial data.
A polished "Your biggest weakness is X" card is dangerous if half the required evidence failed to load.
The state of the data has to influence the state of the interface.
Automated validation got much better
The latest client pass was much more coherent.
Type checks.
Builds.
Tests.
Architecture cleanup.
That gave me more confidence in the package itself.
But I am trying very hard not to repeat the lesson from the previous article.
Automated green does not mean complete.
Real Android behaviour.
Physical screen-reader testing.
Actual browsers.
Real multi-device flows.
Those still need external verification.
The interface can be structurally accessible before it has been fully certified in the environments that matter.
Intelligence finally feels like its own product
That is the part I am happiest about.
It no longer feels like:
The analytics app.
It feels more like:
The place where the preparation gets interpreted.
That is a much stronger identity.
Verbal is where I learn language.
Quant is where I solve.
Intelligence is where the system tries to make sense of the learner those products are observing.
The interface finally started reflecting that job.
I rebuilt the UI because the model deserved a better interface
That is the difference from cosmetic redesign.
I was not bored of the colours.
The old interface was underselling the architecture.
A system built around uncertainty, evidence and action needs hierarchy that makes those ideas obvious.
Otherwise all that thinking remains hidden behind equally weighted cards.
The new design is darker.
Richer.
More colourful.
More deliberate.
But the real improvement is simpler:
It finally knows where to look first.