Wavelength

Inspiration

We're students at LSE, where a typical week means hundreds of pages of dense readings. For me (Lukas), this has always been a real struggle. I deal with attention issues, and being fifteen pages in without absorbing anything isn't an occasional frustration, it's daily. That experience made me look outward and realise how many students face the same thing, or worse.

For students with ADHD and dyslexia, the problem is dramatically amplified. They're every bit as intelligent as their peers, but academic materials are built for neurotypical brains: walls of 12pt Times New Roman, unchanged since 1996. Disability services are overwhelmed with long waits and generic accommodations. The gap between how these students process information and how content is presented is where learning breaks down.

We built Wavelength because this is personal. AI can close that gap, not by doing the learning for students, but by making the material accessible to how their minds actually work.

What it does

Wavelength takes dense academic text and transforms it into neurodivergent-friendly formats. Paste in a paragraph from a textbook or lecture notes, and the AI breaks it into clearly labelled, digestible chunks, simplifies the language without losing academic meaning, and highlights key concepts with plain-language definitions.

The interface is designed from the ground up for accessibility: toggle to a dyslexia-friendly font, adjust spacing and text size, apply colour overlays to reduce visual stress, or have any section read aloud with text-to-speech. A focused study mode presents content one chunk at a time with a progress tracker and gentle recall questions that help students actively engage with the material rather than passively scroll through it.

Critically, the original text is always one toggle away. Wavelength transforms the format, never the substance.

How we built it

We built Wavelength as a two-person team in 2.5 hours using Claude as both our development tool and the AI engine powering the product.

Frontend: React via Claude Artifacts, styled with Tailwind CSS. The interface prioritises calm, accessible design with warm colours, generous spacing, and no visual noise.
AI transformation engine: Claude API (Sonnet) with a carefully crafted system prompt that instructs the model to chunk, simplify, and extract key concepts while preserving academic accuracy. The system prompt enforces strict rules: never remove hedging language, never make uncertain claims sound certain, never invent information.
Accessibility features: OpenDyslexic font toggle, adjustable line height and letter spacing, colour overlay options, and text-to-speech via the Web Speech API.
No backend or database. All state is managed in-memory via React, keeping the architecture simple and the build fast.

David drove the build, prompting Claude to construct features iteratively. Lukas managed sprint timing, quality-tested every feature, wrote the AI system prompt, and prepared the pitch and submission.

Challenges we ran into

Time pressure. 2.5 hours is brutally short. We had to be ruthless about scope, cutting features we wanted (a progress dashboard, an AI study coach chat, export-to-notes) to focus on a polished core experience. Every decision was filtered through one question: "Does this serve the demo?"

Getting the AI simplification right. The system prompt went through several iterations. Early versions either oversimplified (losing important nuance) or barely changed the text. Finding the balance between clearer format and same substance was the hardest design challenge.

Accessibility is harder than it looks. Dyslexia-friendly design isn't just swapping to OpenDyslexic. Colour contrast, spacing, animation choices, loading states: every detail matters when your users are exactly the people most affected by bad UX decisions.

Accomplishments that we're proud of

Built a fully functional, polished prototype in just 2.5 hours with a two-person team
The AI transformation genuinely works: dense academic text goes in, clearly structured and accessible content comes out, with meaning fully preserved
Designed an interface that practises what it preaches, where every design choice (fonts, spacing, colours, animations) was made with neurodivergent users in mind
Kept the ethical framework central to the product, not bolted on as an afterthought. The "original text always one toggle away" decision came from wrestling with real concerns about AI accuracy

What we learned

The gap between "AI tool" and "accessible AI tool" is enormous. Building for neurodivergent users forced us to question every default design choice.
Prompt engineering is product design. The system prompt isn't a technical detail: it's the core of the user experience. How you instruct the AI determines whether the product empowers or patronises.
Ethical thinking isn't a checkbox. The hardest questions, like "what if the AI oversimplifies and a student learns the wrong thing?", shaped our best design decisions.

What's next for Wavelength

LMS integration: connect directly to Canvas, Moodle, and Blackboard so students can transform their assigned readings in one click
Adaptive personalisation: learn from individual usage patterns to surface the right format at the right time
Mobile app: studying happens everywhere, not just at a desk
Multi-language support: neurodivergent students exist in every language
Professional handoff: detect when a student might benefit from human support and connect them to university disability services