Gradient

Gradient is an AR system that empowers all individuals to build any personalized tools through short conversation to generate adaptive tools directly on Raven Resonance’s everyday-wear glasses.

Comment

Gradient

Everyone exists on a gradient. Now everyone can build on one too.
VIEW PROJECT DOCS! (pls click me and view pitch deck)

We had the perfect team combination, a match made in heaven even if none of us had > 1 hackathon experience. So when you have the Raven Glasses that aligns exactly with your ethos, your skill sets, and
everything you've built before, it inspires you to think bigger (if not BETTER). None of that silly hackathon "change the world" talk, but genuinely envisioning what the future could look like for every individual. Creating solutions that stand among the GREATS like our favoirte tool Claude.

But before we could build anything, we had to reframe the entire question.

The hackathon prompt pushed us toward "accessibility tools" and our first
instinct was to build something that helps people with disabilities. But that framing felt wrong. It felt like we'd be building for people rather than
with them. So we stopped, stepped back, and asked ourselves a different
question:

Why are we treating 1.3 billion people like they're identical?

Two developers don't write the same code. Two artists don't create in the same medium. Yet when we build accessibility tools, we slap on clinical
labels—"autism mode," "ADHD mode"—and call it a day. That's simply just categorization.

The reframe that unlocked everything: All people exist on spectrums. All
people are people. Why not let everyone customize a solution based on their
unique position as a human? and adapt to those needs.

This is part of the human experience. The people who would benefit most from
XR are the ones who can build it for themselves.

That's why we made Gradient.

What it does

Gradient is a personalization layer that empowers people with accessibility
needs to dream up any AR application NO CODE, NO STIGMA, and without one-size-fits-all limitations.

Instead of asking "What's your disability?", we ask "How do you experience the world?"

Through natural questions about sensory preferences, cognitive patterns, and
daily challenges, we generate a fully custom AR tool tailored to each user's
unique combination of needs.

We solved a huge stigmatization problem while simultaneously enhancing the
app-building experience for everyone whether you have specific accesibillity needs or not,
whether you're on the spectrum or not. It's the ultimate invisible personalization
layer: a "Build Your Own Agent" experience for AR.

How we built it

This was an insanely complicated workflow involving orchestration of many many many many data servers and llm crunching.

  • AI Orchestrator: A LangGraph-powered pipeline that takes natural
    language input and translates it into AR application specifications
  • Dynamic Generation: Custom UI layouts, features, and interaction
    patterns generated based on user profile
  • Raven Glass Integration: Built specifically for Raven's eye-tracking
    capabilities and Linux-based platform
  • Experience-First Onboarding: A non-clinical question flow that maps user responses to features without stigmatizing language

See our flowchart for the full orchestration architecture.

Challenges we ran into

  1. Getting onboarding right: The onboarding system needed to extract
    intent and data accurately while asking the right questions without making it seem
    like we're fishing for a predetermined solution. Finding that balance took
    multiple iterations.

  2. Prompt engineering at scale: Normal prompting doesn't cut it. We make
    assumptions for you, kind of like every UI component needing features to display its
    data. Orchestrating purpose, pain points, primary features, and visualization while utilizing templates was genuinely difficult.

  3. Documentation digestion: The system is so complex that we needed LLMs
    to help digest our own documentation. Building something this intricate
    required extremely clear documentation just to keep track of everything.

Accomplishments that we're proud of

  • Built onboarding software that truly leverages AI as a tool for human use, the integration layer between human and AI that we can proudly show off
  • Created a working prototype that generates end-to-end in approximately 45
    seconds, producing meaningful AR tools based on user input
  • Fully understood how to integrate with Raven hardware—this is the SaaS layer that can enable so much more for their device
  • Developed architecture and orchestration patterns that would be very
    difficult to replicate in such a short timeframe
  • Designed for an underserved market and opened up a new gap in the space

What we learned

  • People don't want accessibility tools—they want TOOLS. The framing
    matters as much as the functionality.
  • The spectrum is the insight. Treating accessibility as a gradient rather than binary categories unlocks entirely new design possibilities.
  • Scale is wild. We learned to manage 9 Claude context windows
    simultaneously, with each running close to 36 sessions in parallel.

What's next for Gradient

  • Expand the component library: More features and UI components means
    endless possibilities
  • User testing across the spectrum: Test with people across accessibility needs to refine prompts for a broader audience
  • Deeper Raven Glass integration: We want to use the IMU for nodding-based micro-interactions to proceed through flows
  • Partnerships: Explore collaborations with accessibility advocacy
    organizations to bring XR into day-to-day life
  • The vision: Build toward a future where every person can dream up the XR tools they need—designed with utility, accessibility, and human happiness in mind

open to hire! (some of us at least)

Built With

  • claude
  • dedication
  • dreams
  • hope
  • langgraph
  • openai
  • openrouter
  • p5.js
  • passion
  • pydantic
  • python
  • raven-framework
  • support
Share this project:

Updates