AQSA

AQSA (Arabic Quality and Skills Accelerator) was a team entry for the ALLaM Challenge 2024, an international AI competition spanning 17 countries and 177 teams. The platform bundles six Arabic language features behind a multi modal chatbot: I'rab (grammatical parsing), diacritization, dialect to MSA conversion, Mo3gam (Arabic dictionary) search, Holy Quran analysis, and a routing chatbot that accepts text, voice, or image input.

My contributions focused on the Arabic linguistics side. I led the Holy Quran Analysis pipeline that answers grammar, interpretation, and similarity questions over Quranic verses. I built the Quran search engine on Elasticsearch, indexing the 6,236 verse corpus with Arabic-aware analyzers so search could match on roots, diacritics, and verse references with 100 percent retrieval accuracy on our competition benchmark. I also built the Mo3gam dictionary search that looks up Arabic words with context-specific meanings. I contributed alongside teammates on the I'rab parser, diacritization, and dialect conversion features.

The stack: IBM Watsonx hosting the ALLaM foundation model, Python + FastAPI for the features backend, LangChain for prompt routing across the six features, Elasticsearch for the Quran + Mo3gam indexes, a Flutter cross platform app, a Chrome extension (TypeScript), and a TypeScript + Prisma auth backend. All five component repos live under github.com/Allam-Future-Makers.