R U Serious

AI that learns FROM students, not just teaches them. FunLearn activates the Feynman Technique at scale with DigitalOcean Gradient AI — turning every kid into a confident explainer.

Comment

Inspiration

1.5 million students fail board exams in India every year — not because they aren't smart, but because passive learning (reading, memorizing, watching) doesn't work. We were inspired by Richard Feynman's insight: "If you can't explain it simply, you don't understand it well enough." We wanted to build a platform where students teach the AI instead of the other way around — turning every learner into an explainer.

What it does

"R U Serious" is an AI-powered adaptive learning platform with three core engines:

  • Feynman Engine — Students explain topics to Ritty, a curious AI 8-year-old (deployed as a DigitalOcean Gradient AI Agent). He asks probing questions across 5 progressive layers: explain, compress, why-spiral, analogy, and lecture.
  • Story-Based Learning — AI generates engaging stories around any concept with Socratic follow-up dialogue and AI-generated images.
  • Misconception Cascade Tracing (MCT) — A 5-phase Socratic diagnostic that traces misunderstandings to their root cause, then rebuilds understanding from the foundation up.

Every AI response includes an AI-generated educational image (Bria FIBO v2 with Imagen 4.0 fallback), illustration cards, and supports 11 languages including Hindi and Bengali.

How we built it

  • Frontend: React 18 + TypeScript + Vite + Tailwind CSS with Zustand state management, ambient particle effects, and page transitions
  • Backend: FastAPI (Python) with a Provider Factory pattern for swappable AI/image providers
  • AI: DigitalOcean Gradient™ — Serverless Inference (Llama 3.3 70B), Agent Builder (Ritty persona + Guardrails), and Knowledge Base (NCERT curriculum via web crawler)
  • Images: Bria.ai FIBO v2 with automatic Google Imagen 4.0 fallback
  • Infra: DigitalOcean Droplet + Block Storage Volume, fully provisioned via Terraform IaC
  • Data: CSV-based storage on a persistent DO Volume, JWT + bcrypt auth

Challenges we ran into

  • Getting the Feynman Engine's 5-layer progression to feel natural — balancing AI curiosity, confusion detection, and knowledge gap tracking required extensive prompt engineering.
  • Building reliable image generation with automatic fallback — Bria FIBO and Imagen 4.0 have different APIs and response formats, so we built a FallbackImageProvider wrapper.
  • Deploying the full stack (React + FastAPI + Nginx + systemd) on a single Droplet while keeping CSV data persistent across redeployments via a separate Block Storage Volume.
  • Integrating the Gradient AI Agent Builder with a custom persona, guardrails, and a curriculum-grounded Knowledge Base — all provisioned through Terraform.

Accomplishments that we're proud of

  • The Feynman Engine — A genuinely novel approach to AI-assisted learning. Students don't consume content, they produce it.
  • Misconception Cascade Tracing — Goes beyond "right/wrong" to find why a student is wrong, tracing the root cause across prerequisite concepts.
  • Full infrastructure-as-code — One terraform apply provisions the Droplet, Volume, AI Agent, Knowledge Base, and Firewall.
  • Provider Factory — Swapping AI or image providers requires changing one environment variable — zero code changes.
  • 11-language support enforced at the prompt level across every feature.

What we learned

  • The Feynman Technique is incredibly powerful when paired with AI — students who explain concepts develop dramatically deeper understanding than those who just read about them.
  • DigitalOcean's Gradient AI stack (Agent Builder + Knowledge Base + Serverless Inference) provides a complete AI application platform — we didn't need to manage any ML infrastructure.
  • Building educational AI requires careful guardrails — Ritty needed to be curious and probing without being discouraging or overwhelming.
  • Automatic fallback chains for external APIs are essential — no single provider has 100% uptime.

What's next for R U Serious

  • Voice Mode — Full spoken interaction with speech-to-text and text-to-speech so students can explain concepts by talking, not typing.
  • Video Generation — Auto-generate short recap videos from learning sessions using FFmpeg.
  • Team Tournaments — Multiplayer Feynman battles where students compete to explain concepts.
  • Teacher Dashboard — Let teachers see which concepts students struggle with most, powered by MCT diagnostic data.
  • Mobile App — React Native version for offline-first learning in low-connectivity areas.
  • Expanded Curriculum — Ingest more Indian state board and international curricula into the Knowledge Base.

Built With

  • bria-fibo-v2
  • digitalocean-agent-builder
  • digitalocean-droplet
  • digitalocean-gradient-ai
  • digitalocean-knowledge-base
  • digitalocean-volume
  • fastapi
  • llama-3.3-70b
  • python
  • react
  • typescript
Share this project:

Updates