Featured
Arabic RAG Search over Legal Corpus
Hybrid BM25 + dense retrieval pipeline over 2M Arabic legal documents.
Role
Lead AI Engineer
Impact
Cut legal research time from ~40 minutes to under 2 minutes per query.
Year
2024
Retrieval-augmented search over a large Arabic legal corpus. Combined BM25 lexical retrieval with a multilingual dense encoder, reranking with a cross-encoder, and grounded answers with citation spans. Deployed on Azure with a LangChain orchestration layer and Milvus for vector storage.
Stack
RAG · NLP · LangChain · Milvus · LLM