Arabic RAG Leaderboard: A Comprehensive Framework for Evaluating Arabic Language Retrieval Systems
•
15
AI & ML interests
Arabic & bilingual LLM pretraining • Fine-tuning (domain & cultural alignment) • Preference alignment (SFT, DPO, PPO, RLHF/RLAIF) • Reinforcement learning • Evaluation harnesses (Arabic & multilingual) • Efficient RAG (re-ranking & guardrails) • ASR • TTS • Vision-language models (VLMs)
Recent Activity
Articles
Organization Card
We are a team that seeks to connect our people's past with the world's future.
At Navid, we build Arabic-first and multilingual GenAI solutions across domains, laser-focused on delivering real value for people and organizations.
What We Do
- Data pipelines & governance — automated collection, filtering, and curation
- Synthetic data & augmentation — domain-balanced recipes for Arabic tasks
- Training & fine-tuning — SFT, DPO, PPO, RLHF/RLAIF
- Custom LLMs & evals — Arabic/multilingual models with robust evaluation
- Advanced RAG — secure retrievers, re-rankers, hybrid search & guardrails
- Agentic workflows — LangGraph/LangChain multi-agent apps with memory/tools
- Deployment — on-prem, VPC, or managed GPU with full observability
Highlights
We contribute to the Arabic AI ecosystem with:
- No.1 Arabic model - one of the top models for its size trained by Navid.
- First Arabic RAG Leaderboard — open-sourced for the community
- RAG made for Arabic — 1st RAG made for Arabic and English retrieval
datasets
0
None public yet