Neura

Landing Page
On boarding for Context
Main Interface
Choose a topic

Inspiration

Every child learns differently. But education at scale has always been one-size-fits-all. We were inspired by Neal Stephenson's The Primer , a book that adapts to who its reader is, teaching not just facts but thinking itself. We asked: what if we could bring that to every child, everywhere?

What it does

Neura is an adaptive AI tutor for children aged 8-12. Parents input what their child struggled with in school. Neura generates a personalized story lesson where the child's favorite character encounters a problem that naturally teaches the concept they need to understand. Through Socratic questions embedded in the story, children discover answers themselves — building independent thinking, not just memorization.

Two modes:

Parent Mode: Input child's learning gap + interests → AI generates lesson
Child Mode: Read personalized story → Answer guiding questions → Learn by discovery

How we built it

Frontend: React 19, Next.js, Tailwind CSS, Framer Motion (for story text animations)
Backend: Node.js/FastAPI, Gemini 2.5 Flash API
Core: Adaptive prompt engineering that generates story + Socratic questions based on child profile + learning gap
UX: Chalkboard aesthetic (board texture, chalk colors), voice narration, smooth animations, progress tracking

Story generation flow: Parent input → Gemini prompt (context-aware) → JSON with 5 scenes + 4 questions → Child interface renders story progressively with fade-in animations and voice-over.

Challenges we ran into

Token constraints: Initial prompt was too verbose (8192 tokens). Optimized to 2500 to fit within rate limits.
Prompt quality: Getting the AI to actually teach Socratically (guide, don't answer) required careful prompt engineering and iterative testing.
Story coherence: Ensuring each question flows naturally from the narrative (not bolted on) while maintaining pedagogy.
API reliability: Multiple fallback chains (Gemini → Groq → OpenRouter) to handle rate limits during hackathon crunch.

Accomplishments we're proud of

End-to-end MVP: Working from onboarding → story generation → child interaction in 16 hours
Pedagogically sound: Actually teaches through questioning, not just entertainment
Personalization at scale: Every lesson is unique to the child (name, age, interests, learning style, specific struggle)
Polish: Beautiful UX with animations, voice narration, confetti celebrations — feels like a real product
Feasibility: Low token cost, works with free API tiers, deployable today

What we learned

Pedagogy + AI > just AI alone. The prompt engineering for Socratic method was harder than the code.
Simplicity wins. Story-only mode (text + questions) outperformed complex chalkboard visualizations for both feasibility and learning.
Personalization is the differentiator. Not flashy features — knowing the child's name and interests makes lessons stick.
Hackathon scope matters. We cut 70% of planned features and shipped something that works instead of something ambitious that breaks.

What's next for Neura

Voice input for younger children (8-9 year-olds who struggle with typing)
Game mode with interactive challenges alongside stories
Multi-subject expansion (currently math focus, add science, languages, logic)
Parent dashboard with progress tracking, learning gaps identified, recommendations
Fine-tuned model for production (custom Gemini tuning for teacher-quality responses)
Offline mode for low-connectivity regions
Localization (Arabic, Spanish, French, Mandarin — built for global learners)