EchoForge
A bilingual (zh/en) audio content publishing platform. Upload audio files, let the pipeline transcribe and process them with AI, then publish polished articles in both languages.
What it does
- Upload an audio file via the admin panel
- The pipeline transcribes the audio, uses a local LLM to refine and translate the content, and writes the result back
- Review and publish the post — it appears on the public-facing site in both Chinese and English
Tech stack
| Layer | Technology |
|---|---|
| Frontend / API | Next.js (App Router), deployed on Vercel |
| Database | Neon PostgreSQL |
| File storage | Vercel Blob |
| Auth | NextAuth.js + Google OAuth |
| AI pipeline | Ollama (Qwen2.5 32B), runs locally |
Architecture
The pipeline runs on a local machine and polls the database every 30 seconds for new jobs. Vercel and the pipeline never communicate directly — Neon DB is the only bridge. This keeps the local LLM unexposed and prevents Vercel functions from timing out on long inference tasks.
See ARCHITECTURE.md for the full system diagram.
Environment variables
| Variable | Description |
|---|---|
DATABASE_URL | Neon PostgreSQL connection string |
AUTH_GOOGLE_ID | Google OAuth client ID |
AUTH_GOOGLE_SECRET | Google OAuth client secret |
AUTH_SECRET | NextAuth.js secret |
BLOB_READ_WRITE_TOKEN | Vercel Blob token |
ALLOWED_EMAIL | Admin email address (only this account can log in) |
Development
npm install
npm run dev
Open http://localhost:3000.