convez.ai
InicialPluginScore · de 0 a 100

ReadU — adaptive AI reading comprehension for middle schoolers, personalized by skill & interest. Built with Next.js + Claude (Opus 4.8).

Instalação

$npx convez-ai willfeldman/readu

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

README

A tradução para o português deste README ainda está na fila.

ReadU 📖 — Reading that reads you

An adaptive AI reading-comprehension platform for middle schoolers. ReadU writes reading passages and comprehension questions personalized to each student's reading level and their interests, then adapts — passage by passage — to target the comprehension skills they most need to grow.

Built for Build Day. Powered by Claude (Opus 4.8).

▶ Live demo — readu-five.vercel.app

Deploy with Vercel

One click deploys ReadU in library mode — no setup, works instantly. To enable live, per-session passage generation, add an ANTHROPIC_API_KEY environment variable in your Vercel project afterward (get one at console.anthropic.com).


✨ What it does

  1. Onboarding — the reader tells ReadU their name, grade, and a few interests (space, sports, mystery, gaming, ocean life… 12 themes).
  2. Baseline reading — a short passage + 5 questions calibrates a starting reading level (grades 3–9).
  3. Adaptive sessions — every session, Claude writes a brand-new passage on a topic the student loves, at their current level, with questions weighted toward their weakest comprehension skills.
  4. It adapts — score high and the text gets richer; struggle and it eases back. The eight tracked skills (main idea, inference, vocabulary, cause & effect, sequencing, author's purpose, key details, figurative language) update after every answer.
  5. Stays motivating — XP, ranks, day streaks, badges, a live skill dashboard, and reading history keep kids coming back.

The core thesis: personalization drives engagement, and engagement drives literacy. A kid who can't put down a story about their favorite game is a kid who's reading.


🧠 How the AI works

Every passage and question set is generated by Claude (Opus 4.8) in two layers:

  • Live generation (/api/passage → lib/generate.ts): on each session, the reader's interests, current level, recent topics, and lowest-mastery skills are sent to claude-opus-4-8 using structured outputs (output_config.format with a strict JSON schema), so the response is always a valid passage + 5 questions + a vocabulary word bank. No key required to try the app — see below.
  • Built-in library (data/seed-passages.json): 36 passages (12 interests × 3 grade bands) pre-generated by Opus 4.8 so ReadU works fully offline and is instantly demo-able. If no ANTHROPIC_API_KEY is set (or a live call fails), ReadU seamlessly serves a matching library passage and tells you so.

The adaptive engine lives in lib/adaptive.ts (pure, dependency-free): per-skill mastery is an exponentially-weighted moving average; reading level nudges up/down based on session accuracy; and nextFocusSkills() feeds the weakest skills back into the next generation request — closing the loop.


🚀 Run it

npm install

# (optional but recommended) enable LIVE, personalized generation:
cp .env.example .env.local
#  → put your Anthropic API key in .env.local

npm run dev
# open http://localhost:3000
  • With an ANTHROPIC_API_KEY: Claude writes a fresh passage every session.
  • Without a key: ReadU runs entirely on its built-in Opus 4.8 library — great for a no-setup demo. Everything else (adaptivity, skills, badges, XP) works identically.

Progress is stored in localStorage (no database, no login) — reset any time from the dashboard.


🛠 Tech stack

  • Next.js 14 (App Router) + TypeScript
  • Tailwind CSS — custom design system (Bricolage Grotesque + Plus Jakarta Sans)
  • Framer Motion — transitions, confetti, micro-interactions
  • Anthropic SDK — claude-opus-4-8 with structured outputs
  • lucide-react — icons

📁 Structure

app/
  page.tsx              # screen state machine (onboarding → baseline → dashboard → session)
  api/passage/route.ts  # POST → live Opus 4.8 generation, library fallback
  layout.tsx, globals.css
components/              # Onboarding, ReadingSession, PassageReader, QuestionCard,
                        # ResultsView, Dashboard, LevelMeter, SkillBar, Confetti, …
lib/
  generate.ts           # Claude (Opus 4.8) passage generation + JSON schema
  content.ts            # built-in library selection + fallback
  adaptive.ts           # reading-level + per-skill mastery engine
  skills.ts             # skill / interest / badge catalogs
  store.ts              # localStorage profile hook
  types.ts              # shared types
data/
  seed-passages.json    # 36 Opus 4.8-generated passages (offline/demo content)

License

MIT © Will Feldman — free to use, learn from, and build on.

Made for Build Day with Claude. 🎓

Ver o original no GitHub

Bom para

Relacionados