Inspiration

Every classroom is filled with students who learn differently, yet educational content is still overwhelmingly delivered in a single format. Students with ADHD, dyslexia, auditory processing disorders, visual impairments, or simply different learning preferences are often expected to adapt to the material instead of the material adapting to them.

Most existing platforms ask students how they prefer to learn. We believed there was a better approach.

Triad determines a student's optimal learning modality through a short diagnostic assessment, then automatically delivers lessons in the format where they perform best—text, audio, or visual. Rather than relying on self-reported preferences, the platform continuously measures learning outcomes and adapts instruction accordingly.

What it does

Triad is an AI-powered adaptive learning platform for both teachers and students.

For Teachers

Upload existing course materials (PDFs, documents, presentations)
Automatically generate:
- concise lesson summaries
- practice quizzes
- audio narration
- visual learning content
Monitor class-wide and individual student performance through an analytics dashboard
View each student's recommended learning modality and assessment history

For Students

Students first complete a diagnostic assessment that evaluates learning performance across multiple content modalities.

Based on their results, Triad automatically presents future lessons in the format where they demonstrate the highest comprehension:

📖 Text-first learning
🎧 Audio narration
🎨 Visual explanations

Students can also interact with an AI tutor through voice or text, complete quizzes, and track their progress over time.

The platform also supports predefined accessibility accommodations that can bypass the diagnostic process when required (for example, mandatory audio narration or caption requirements).

How we built it

Our architecture consists of a React frontend backed by a Next.js API layer.

Frontend

React
Vite

Backend

Next.js 16 App Router API Routes

Infrastructure

Supabase
- PostgreSQL database
- File storage
Upstash Redis
- Caching
- Student state management

AI Stack

Anthropic Claude (Opus 4.8)
- Generates multimodal educational content
OpenAI GPT-4o mini
- Powers the conversational AI tutor
Deepgram
- Text-to-Speech
- Speech-to-Text

Teachers upload instructional material, which is processed through our AI content pipeline to generate multimodal learning resources. Student assessment results are stored in Supabase, cached with Redis for fast retrieval, and used to personalize future lessons automatically.

Challenges we ran into

One of our biggest challenges was designing a system that actually measures how students learn best instead of simply asking for their preference.

We also had to coordinate multiple AI services while maintaining a smooth user experience. This involved generating multimodal content, supporting conversational tutoring, and integrating real-time speech capabilities without making the platform feel fragmented.

Another challenge was designing for accessibility from the beginning. We wanted accommodations to be treated as first-class features rather than afterthoughts, allowing students with predefined accessibility needs to receive appropriate content immediately.

Accomplishments that we're proud of

Built an end-to-end adaptive learning platform in a hackathon timeframe.
Created an objective diagnostic assessment that recommends learning modalities based on demonstrated performance.
Successfully integrated multiple AI providers into a single cohesive workflow.
Designed separate teacher and student experiences with personalized dashboards.
Implemented voice interaction, multimodal lesson generation, AI tutoring, and learning analytics in one unified platform.

What we learned

Building Triad reinforced that personalization is more than simply adding AI—it requires thoughtful measurement, feedback loops, and accessibility-first design.

We also learned how powerful modern AI models become when each specializes in a specific role. Using Claude for educational content generation, GPT for conversational tutoring, and Deepgram for speech allowed us to leverage each model's strengths instead of forcing one model to do everything.

Finally, we learned the importance of designing educational technology around measurable learning outcomes rather than assumptions about how students think they learn.

What's next for Triad – Learn Your Way

Our next goal is to make Triad adaptive over time instead of relying solely on an initial assessment.

Future improvements include:

Continuous modality adjustment as students learn
Personalized study plans generated from historical performance
Teacher recommendations for intervention when students begin struggling
LMS integrations (Canvas, Google Classroom, Blackboard)
More accessibility options for diverse learning needs
Longitudinal analytics showing how learning preferences evolve over an academic year

Ultimately, we envision Triad becoming an AI-powered learning companion that continuously adapts alongside every student, ensuring educational content is delivered in the way they learn best—not the way it was originally created.