Ludwig.guru

Most writing tools try to answer a binary question: is this sentence correct? Ludwig helps users ask more useful questions: does it sound natural, how do credible writers express the same idea and which alternative fits this context better?

A sentence can be grammatically valid and still sound unnatural. Two phrases may mean almost the same thing while carrying different levels of formality, confidence or emotional weight.

Ludwig connected suggestions to real examples and visible sources. Any redesign needed to protect that behaviour. Simplifying the experience could not mean removing the evidence that made it useful.

People did not only want correction. They wanted evidence.

Ludwig began with one distinctive interaction: enter an expression, search for it and inspect how similar language appears in credible sources. As it expanded, search, comparison, translation, correction and rewriting were added around that behaviour.

Each capability was useful by itself. Together they made users translate a broad intention—‘I want to write this better’—into Ludwig’s internal divisions. The interface treated actions as destinations. Users experienced them as consecutive moments in one flow.

Results also had depth without a clear hierarchy, platform patterns had diverged and every new feature risked introducing another local design solution.

• Feature-led navigation
• Competing result blocks
• Diverging platform patterns
• A design foundation that was hard to scale

Cleaner layouts, new typography and updated controls were necessary, but they did not answer the central question: how could Ludwig grow into a broader writing platform without losing its contextual search experience?

The redesign needed to preserve the familiar search behaviour while creating clearer relationships between the growing capabilities.

The aim was not to replace the existing mental model overnight. It was to give the product enough structure to evolve without becoming a menu of loosely connected writing tools.

Before redesigning screens, I mapped how people entered the product, submitted a query, interpreted results, explored alternatives, continued working, encountered limits and moved between anonymous, free and paid states.

The map covered search and input states, result flows, alternative writing actions, account states, product boundaries, empty, loading and error states, responsive behaviour and desktop and mobile interactions.

Many interface problems were symptoms of one structural issue: the experience had been designed around individual capabilities rather than a continuous writing journey.

I reorganised the experience around what users were trying to achieve rather than the technical divisions of the product.

The actions remained distinct because search, correction, comparison, translation and longer-form writing do not produce identical results. But they began to share input patterns, navigation, terminology and controls.

This created continuity without pretending every writing problem was the same. A new capability no longer needed users to understand another miniature product.

Ludwig already provided examples, alternatives and contextual evidence. The design challenge was to show what that information meant for the user’s own sentence.

I introduced five levels: the primary response, suggested improvements, real-world examples, sources and supporting information, and possible next actions.

A user looking for a quick answer could understand the main result immediately. A researcher, translator or professional could inspect examples and sources in depth. Information remained available, but no longer competed at the same visual level.

Language does not always have one best answer. A phrase may be common in academic writing and awkward in conversation. Two alternatives can both work while expressing different intentions.

The redesign preserved the visible relationship between suggestions, usage examples, context and sources instead of asking users to trust a mysterious correction.

The product’s job was not always to say ‘use this sentence.’ Sometimes the more useful response was to present the strongest options and the evidence needed to choose between them.

Confidence came from evidence the user could inspect, not from hiding uncertainty.

Different writing tools could easily have introduced separate entry points, forms and rules. I designed the input as a shared starting point across the experience.

Users could begin with the language already in front of them, then compare it, translate it or improve it without copying the sentence into separate tools.

Results, alternatives and related actions became connected parts of the same process. Users could explore different versions of an idea without losing the sentence that started the task.

The redesign covered web, desktop and mobile. The objective was behavioural continuity rather than visual duplication.

On larger screens, the interface supported denser information and side-by-side comparison. On mobile, the main answer appeared first while evidence and secondary actions were revealed progressively.

The same concepts—query entry, result priority, alternatives, source access and continuation—remained recognisable across devices, so users did not need to reconstruct their understanding each time.

Mobile users may check a sentence inside an email, review a translation during a conversation or make a quick decision between other tasks.

The experience prioritised immediate interpretation, clear alternatives, progressive evidence, reduced visual competition and predictable continuation actions.

The redesign did not remove the depth of Ludwig’s results. It changed when that depth appeared.

I built reusable foundations and components for typography, spacing, colour, inputs, navigation, controls, result cards, system states, responsive behaviour and page layouts.

The system was designed around behaviour rather than finished screens. Components needed to handle different result types, variable content length, partial information, multiple interaction states and future writing capabilities.

A language product may show one short sentence or several long examples. A source title may fit one line or read like an academic committee was paid by the word. A design that only works with ideal placeholder text is not a product system. It is a screenshot.

Ludwig had an established user base. A complete reinvention might have produced a more dramatic before-and-after image, but it would also have increased product risk.

The redesign retained the central role of search, the evidence-based result model, real examples and the connection between suggestions and sources. It changed the hierarchy, predictability, relationships between tools and continuity across devices.

Speed had to coexist with depth, consistency with platform context, familiarity with change and feature visibility with interface simplicity.

• Speed vs depth
• Consistency vs platform context
• Familiarity vs reinvention
• Feature visibility vs interface simplicity

The redesign created a unified foundation across web, desktop and mobile: a clearer relationship between writing tools, a more understandable path from query to decision, a stronger hierarchy for complex results and a scalable component system.

There were no reliable baseline measurements for precise claims about faster completion or improved writing outcomes. The impact was visible in the product structure: disconnected writing actions moved into a shared model and the platforms began to behave as parts of the same system.

The hardest work was not drawing screens. It was deciding what should remain familiar, what needed to change and which patterns could support a product that was still evolving.

A coherent product is one in which users understand how the pieces help them complete the same job.