LLM & RAG Engineer · Multilingual NLP (Arabic + Urdu)
I'm an MSc NLP student at MBZUAI and a data-team intern at the Institute of Foundation Models, where I work on Arabic LLM development. I build LLM and RAG systems — retrieval, fine-tuning, and agent workflows — with a focus on Arabic and Urdu, two languages most models still handle poorly.
📍 Abu Dhabi, UAE · Open to LLM/RAG engineering roles and freelance projects
- RAG — chunking, embeddings, vector search, and grounded answers that cite their source
- Fine-tuning — LoRA / QLoRA and instruction tuning, quantized to train on a single GPU
- Agents — multi-agent LangGraph pipelines where agents check each other's output
- Multilingual NLP — Arabic and Urdu, including low-resource and dialectal data
| Project | What it does |
|---|---|
| InvestAI | Multi-agent advisor that turns a raw bank statement into a constraint-aware investment portfolio |
| CodeCop | Detects machine-generated code with a QLoRA-fine-tuned transformer, benchmarked against prompting |
| AgriVerse | Pairs an LLM crop advisor with LSTM price forecasting across 200 stations |
- SciDER
— autonomous multi-agent system for end-to-end scientific research. Co-author and a core contributor. Live demo · Project page · Paper
- UrduMMLU: A Massive Multitask Benchmark for Urdu Language Understanding — arXiv preprint, 2026. A 26k-question, 26-subject benchmark for evaluating LLMs on Urdu (with P. Nakov et al.). Paper
- SciDER: Scientific Data-centric End-to-end Researcher — arXiv preprint, 2026. A multi-agent system for automating data-centric scientific research workflows (with P. Nakov et al.). Paper · Code · Demo
- Oath Breakers at SemEval-2025 Task 6: PromiseEval — SemEval-2025 (ACL). Placed 2nd on the English leaderboard. Paper
- Fine-tuned LLM Approach to Scientific Text Simplification — CLEF 2024, SimpleText track (CEUR). Paper
- Analysis of Algorithmic Design Techniques for Seam Carving — arXiv, 2024. Paper
Full list on Google Scholar.