Context and architecture
Developed for the Ministry of Higher Education and Scientific Research to make official Arabic documents searchable through natural-language questions.
The pipeline retrieves source passages, grounds answers in the underlying material, and returns document references through a full web interface.
Core capabilities
- Arabic semantic search
- Document-grounded answers
- Source-aware retrieval
- Web interface for search and administration
Engineering challenge
Arabic morphology and orthographic variation weakened off-the-shelf retrieval. A dedicated preprocessing layer was added before embedding and search.
Outcome
Indexed an estimated 1,200+ documents, reached estimated sub-two-second average response latency, and achieved estimated answer relevance above 85% on held-out queries.