cv

Research experience, education, interests, and publications.

Contact Information

Name Atif Quamar
Professional Title Visiting Researcher at MBZUAI
Email mohammad.atif@mbzuai.ac.ae
Website https://www.atifquamar.com

Summary

I like to understand how foundation models reason, how their reasoning abilities emerge through training, and how we can make that reasoning more efficient and trustworthy. Much of my work spans language and vision, with a focus on training and aligning models to be safe and aligned with human values.

Education

  • M.Sc in Machine Learning

    Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), Abu Dhabi, UAE

    Machine Learning

  • B.Tech in Computer Science and Biosciences

    Indraprastha Institute of Information Technology (IIIT), Delhi, India

    Computer Science and Computational Biology

Experience

  • Visiting Researcher

    Mohamed bin Zayed University of Artificial Intelligence

    Developing training and inference-time methods to improve reasoning in language models, including reinforcement learning, self-consistency, and test-time inference strategies, with applications to mathematical reasoning and coding while reducing unreliable reasoning paths.

  • Research Intern

    University of Virginia

    Developed a drift-resilient memory framework for code-execution agents using KL-constrained adapter updates, mitigating embedding distribution shifts during online learning to reduce unsafe code generation without compromising task success rates.

  • Research Intern

    University of California San Diego

    Proposed a multimodal Chain-of-Thought framework that interleaves text with latent visual representations in VLMs, with a two-stage SFT + RL training setup for modality switching, achieving strong gains across 11 multimodal reasoning benchmarks.

  • Research Intern

    Purdue University

    Studied Bayesian sampling and inference-time alignment methods for language models, improving response quality and steering outputs toward harmlessness and positive sentiment.

  • Co-Founder

    Insituate

    Founded Insituate and grew it to $250K ARR, building on-premise AI agents using open-source LLMs and RAG systems for enterprise use cases; deployments included Singapore Judiciary, Indian High Courts, Mizuho Bank, and PNC Bank.

Publications

Interests

Research: Multimodal Reasoning, Efficient Training, AI Alignment and Safety, Reinforcement Learning