Public backend project
Arabic NLP Toolkit API
A lightweight FastAPI service for Arabic normalization, transliteration, tokenization and spelling out numeric amounts.
The problem
Arabic applications need predictable text cleanup and properly formatted numeric amounts without loading a large model.
My contribution
Implemented reusable Arabic text utilities and exposed them through validated API endpoints.
Engineering decisions
- Keep the processing rule-based and dependency-light.
- Provide reversible Buckwalter transliteration and configurable normalization.
- Treat dialect detection as a lexical heuristic, with a fallback when markers are sparse.
Results & evidence
- Seven existing core tests pass for normalization, transliteration, number grammar, dialect markers and token statistics.
- Includes container and deployment configuration; a public hosted service has not been verified.
