EchoWrite

EchoWrite is a production-style MVP for a back-translation writing trainer. It helps users practice English writing by reconstructing English from their own Chinese translation, then reviewing structured coaching feedback focused on writing quality rather than literal translation.

Screenshots

Home

Practice flow

Attempt feedback

Mistake review

Passage library

Stack

• Next.js App Router
• TypeScript
• Tailwind CSS
• shadcn-style component layer
• Prisma + SQLite
• Zod
• OpenAI server-side evaluator with a mock fallback

Architecture

• app/: App Router pages, server actions, and the /api/evaluate route.
• components/: UI primitives, page scaffolding, forms, and result renderers.
• lib/db.ts: Prisma client singleton.
• lib/evaluator/: evaluator abstraction, mock heuristics, OpenAI prompt, and service entrypoint.
• lib/validators/: Zod schemas for passages, attempts, and evaluation payloads.
• prisma/: schema plus seed script with demo passages and sample attempts.
• scripts/: local helper script for resetting the database.

MVP Scope

Implemented:

• Home page with clear entry points
• New practice flow
  • Step A: source English to Chinese
  • Step B: Chinese to back-translated English
  • Step C: scored results with structured feedback and text diff
• Passage CRUD
• Attempt history with filters
• Mistake aggregation by category
• Mock evaluator for local development
• OpenAI evaluator behind EVALUATOR_MODE=openai
• Seeded demo data

Not included:

• Auth
• Collaboration
• Payment
• Background jobs
• Advanced analytics dashboards

Local Setup

1. Install dependencies:

npm install

2. Copy the environment file:

cp .env.example .env

3. Create the SQLite database and seed demo content:

npm run db:push
npm run db:seed

4. Start the dev server:

npm run dev

Open http://localhost:3000.

Evaluator Modes

Mock mode

Default local mode:

EVALUATOR_MODE=mock

This returns deterministic structured feedback without an API key.

OpenAI mode

Set these in .env:

EVALUATOR_MODE=openai
OPENAI_API_KEY=your_key_here
OPENAI_MODEL=gpt-5.2

The evaluator is called server-side and validates model output with Zod. If the model response is malformed or the request fails, the app falls back to mock evaluation so the MVP remains usable.

Useful Commands

npm run dev
npm run lint
npm run typecheck
npm run db:generate
npm run db:push
npm run db:seed
npm run db:reset

Notes

• Tags are stored as a JSON array in SQLite for MVP simplicity.
• Attempts keep a snapshot of evaluationJson plus denormalized numeric scores for filtering.
• Mistakes are materialized into a dedicated table so recurring patterns can be aggregated quickly.