Inspiration
NeuroOrbit AI started from a simple problem: educational content can be understandable but still difficult to start.
For a student facing dense instructions or a multi-step assignment, the problem is not necessarily a lack of knowledge. It can be knowing what to do first, understanding which requirements matter most, or dealing with an interface that presents too much information at once.
We wanted to explore a different approach: instead of building another general-purpose AI chatbot, we designed an accessibility-first learning workspace where students can adapt how educational content is presented according to their own preferences.
The project was also inspired by our work at Space Dogs and our interest in making technical and scientific education more accessible. Although our first demonstrations include space-science content, NeuroOrbit AI is designed to work with ordinary educational material as well.
What it does
NeuroOrbit AI transforms lesson text into a structured Mission Path.
A student provides lesson or assignment text and selects learning preferences such as:
- Fewer words
- Smaller steps
- More examples
- Visual explanations
The system is designed to transform that input into a structured path containing:
- An objective
- Simplified requirements
- A first actionable step
- Ordered steps
- A completion checklist
The student can then work through the steps progressively instead of having to process the entire assignment at once.
The interface also includes accessibility-oriented controls for adjusting text size, line spacing, reducing distractions, reading content aloud, and navigating the application with the keyboard.
A refinement flow allows the student to request changes such as making the first step smaller, using fewer words, showing an example, or explaining the material differently.
The central principle is that adaptation should change the presentation of the task, not silently remove its essential requirements.
How we built it
We deliberately kept the architecture small so that we could concentrate on the interaction between the learner, the AI adaptation layer, and the accessible interface.
The frontend uses React and Vite, with CSS-based presentation and browser capabilities such as the Web Speech API and local storage. The documented frontend is organized around components such as the lesson input, preference controls, Mission Path renderer, accessibility controls, and feedback/refinement interactions.
The backend uses Python, FastAPI, and Pydantic. The architecture defines a small API surface consisting of /health, /api/adapt, and /api/refine. The adaptation and refinement contracts are represented through structured Pydantic models, with MissionPath acting as the shared structure between the adaptation and refinement flows.
The intended data flow is:
Lesson Text
↓
Learning Preferences
↓
React / Vite
↓
POST /api/adapt
↓
FastAPI
↓
AI adaptation layer
↓
Validated MissionPath
↓
Accessible frontend
We also kept accessibility preferences separate from AI preferences. AI preferences affect how the learning content is adapted, while presentation preferences such as text size, spacing, and low-distraction mode belong to the frontend.
The current repository already contains the foundation for this architecture, including CORS configuration, the MissionPath schema, adaptation and refinement routes, accessibility controls, sample lessons, and frontend-to-backend communication.
Challenges we ran into
One of our biggest challenges was avoiding the temptation to build too much.
Our previous projects at Space Dogs often involve engineering-oriented systems with APIs, telemetry, dashboards, simulations, and other technical components. NeuroOrbit AI required a different approach. The primary challenge was designing around a learner's experience rather than around technical complexity.
We also had to separate several concepts that initially seemed similar: AI adaptation, accessibility preferences, and interface design. We learned that not every user preference needs to become an AI parameter or a backend feature.
Another challenge was maintaining a stable structured contract between the frontend and backend. Instead of allowing the AI to return arbitrary conversational text, we designed the system around a predictable Mission Path structure that the frontend can render consistently.
Finally, accessibility required more than visual styling. Keyboard navigation, visible focus, readable text, adjustable presentation, reduced motion, and read-aloud functionality all affect how the application actually behaves.
Accomplishments that we're proud of
We are proud that NeuroOrbit AI became a concrete product architecture rather than just an AI concept.
We established a working foundation connecting a React/Vite interface to a FastAPI backend and defined a structured Mission Path model that can be shared across adaptation and refinement workflows.
We also built the project around accessibility from the beginning instead of treating accessibility as a final visual polish step. The frontend includes controls for text size, line spacing, read-aloud functionality, low-distraction mode, and visible keyboard focus. Another accomplishment was creating a demonstration that connects general education with our background in space technology. One of the included sample lessons explores how satellite imagery can be used to identify urban heat islands, while another uses a general water-cycle assignment.
Most importantly, the project gave us a foundation that can continue beyond the hackathon instead of becoming a one-week prototype that is immediately discarded.
What we learned
We learned that meaningful educational AI is not necessarily about generating more content.
The more important question is often:
What does the learner need to do next?
That changed how we approached the product. Instead of designing a chatbot that continuously produces explanations, we focused on structured adaptation, predictable steps, and learner-controlled presentation.
We also learned that accessibility is an architectural concern. Some adaptations belong in the AI layer, while others belong entirely in the frontend. Keeping those responsibilities separate makes the system easier to maintain and gives users more direct control.
From a technical perspective, we learned the value of establishing the frontend-backend contract before adding unnecessary complexity. A small structured API makes it possible to change the AI implementation later without rebuilding the entire interface.
Most importantly, NeuroOrbit AI changed our perspective on how the engineering experience we have developed through Space Dogs can be applied outside aerospace. The same discipline we use to break down complex engineering problems can also be applied to breaking down complex learning tasks.
What's next for NeuroOrbit AI
The hackathon version is only the beginning.
Our next priority is to replace the prototype adaptation logic with a robust AI adaptation service capable of analyzing real educational material while preserving essential requirements and producing consistently validated Mission Paths.
We also want to expand the refinement system so that learners can continuously adjust an existing Mission Path rather than restarting the adaptation process.
Longer term, we plan to explore:
- More sophisticated AI adaptation and validation
- Better preservation of original assignment requirements
- Visual and multimodal learning structures
- More advanced accessibility personalization
- Persistent learning-path history
- Teacher and educator workflows
- Anonymous usability and feedback analysis
- Integration with educational resources
- Specialized support for technical and scientific education
- Space-science learning modules developed through Space Dogs
Our long-term vision is not to build an AI that decides how every student should learn. It is to build a flexible learning workspace where students can control how complex information is presented to them.
NeuroOrbit AI will continue as an independent project after the hackathon, with the goal of developing the prototype into a more reliable, evidence-driven accessibility and educational technology platform.
Built With
- accessibility
- ai
- artificial-intelligence
- assistive-technology
- css
- edtech
- fastapi
- for
- generative-ai
- javascript
- natural-language-processing
- pydantic
- python
- react
- vite
- web-speech-api
Log in or sign up for Devpost to join the conversation.