← BackProject · 05
AI Novel Translator
Full-stack · Node.js · OpenAIGITHUB ↗
- The problem: reading translated web novels is fragmented. Inconsistent character name rendering across chapters, multiple translators with different conventions, and no single place to follow a series. This platform centralizes it.
- Translation memory stores 500+ terms (character names, place names, genre-specific vocabulary) and intercepts OpenAI API calls to replace model-generated translations with authoritative mappings. The model handles prose; the memory enforces consistency. Improved translation accuracy by ~40%.
- Async translation pipeline processes chapters in parallel using a job queue, decoupling the write path from the OpenAI API round-trip. Throughput improved by 60% over synchronous request/response.
- PostgreSQL stores the novel catalog, chapter index, and translation memory. A dedicated status table tracks translation state per chapter so interrupted jobs resume from the last completed section, with no re-processing and no duplicate API calls.
- Express.js API layer handles 1,000+ concurrent requests; rate limiting and retry logic with exponential backoff ensure OpenAI API quota is never exceeded under sustained load.
- Reader interface with chapter pagination, reading progress persistence per user, and a term glossary sidebar that surfaces the translation memory in context.
Tech Stack
NODE.JS
NODE.JSEXPRESS.JS
EXPRESS.JSPOSTGRESQL
POSTGRESQL