Salah Alzubi

Salaheddin Alzubi

AI Researcher @ Sentient Foundation

Hello! I'm an AI researcher @ Sentient building multi-agent systems.

About Me

I am an AI researcher that's broadly interested in exploring the genrealization capabilties of ML models. I've been fortunate enough to have worked at top labs in academia and industry on a diverse range of problems: from pure research roles to pure engineering roles and roles in-between. I am particularly interested in building impactful systems with LLM's and MLLM's. I am currently working on building open-source multi-agent systems, particularly in search.

I am also interested in any impactful applications of ML related to Arabic/MENA region. Besides work, I enjoy traveling, lifting weights and cooking.

Work Experience

Nov 2024 - Present

AI Researcher

Sentient Foundation

  • Designed and productionized an agentic web-powered LLM service to answer general search queries and complex multi-step queries related to cryptocurrencies. Sole author of the Github repo with 3k+ stars (Publication in repo)
Jul 2023 - Oct 2024

Research Scientist

DAIMON Labs

  • Designed and proto-typed the first production-grade real-time (<800ms) video-to-video pipeline that aims to replicate a FaceTime call with your AI friend
  • Developed and productionized large language models that serve as your personal companion; this includes, but is not limited to: Data collection, Pre-training/Fine-tuning, Optimization, Retrieval Augmented Generation (RAG) and effective evaluation

Selected Publications

Open Deep Search: Democratizing Search with Open-source Reasoning Agents

S Alzubi, C Brooks, P Chiniya, E Contente, C von Gerlach, L Irwin, Y Jiang, A Kaz...

arXiv 2025

We introduce Open Deep Search (ODS) to close the increasing gap between proprietary search AI solutions and their open-source counterparts. ODS augments open-source LLMs with reasoning agents that can use web search tools to answer queries, achieving state-of-the-art performance on SimpleQA and FRAMES benchmarks...

Foundational autoraters: Taming large language models for better automatic evaluation

T Vu, K Krishna, S Alzubi, C Tar, M Faruqui, YH Sung

EMNLP 2024

We present a novel approach to automatic evaluation using large language models, introducing foundational autoraters that significantly improve the reliability and efficiency of natural language generation assessment...

Meta-adapters: Parameter efficient few-shot fine-tuning through meta-learning

T Bansal, S Alzubi, T Wang, JY Lee, A McCallum

AutoML'22

This paper introduces meta-adapters, a novel approach combining meta-learning with parameter-efficient fine-tuning to enable rapid adaptation of large language models with minimal computational overhead...

aiXplain at Arabic Hate Speech 2022: An Ensemble Based Approach to Detecting Offensive Tweets

S Alzu'bi, T Castro Ferreira, L Pavanelli, M Al-Badrashiny

OSACT5 @ LREC'22

We present an ensemble-based approach for detecting hate speech in Arabic social media content, combining multiple deep learning models with linguistic features to achieve state-of-the-art performance...

Contact Me

I'm always open to discussing research collaborations, AI projects, or potential opportunities. Feel free to reach out!