I am a Research Engineer at MBZUAI, working with Dr. Nils Lukas on secure, private, and strategic machine learning. My research focuses on language-model agents that rely on retrieval, memory, and tools—systems in which many critical failures arise from the surrounding pipeline rather than the base model itself.

Sarim Hashmi

News

  • Jun 2026 NEW Won the Cursor x Evoost AI Hackathon 2026 with Hakim AI — a real-time talking AI avatar for grounded decision intelligence over Abu Dhabi proptech data.
  • Apr 2026 NEW CASS (Nvidia → AMD transpilation) accepted at the ACL 2026 Main Conference.
  • Mar 2026 NEW Submitted PERSUADE — a debate arena for auditing persuasion in LLMs.
  • Mar 2026 NEW Submitted work on robust and calibrated detection of authentic multimedia content.
  • Oct 2025 Received the UAE Golden Visa, awarded for outstanding academic achievement.
  • Aug 2025 Secured 3rd place at MICCAI BraTS 2025 — SSA Track.
  • Jul 2025 Started as Research Engineer at MBZUAI with Dr. Nils Lukas, building Etihad Airways' conversational agent.
  • May 2025 Guaranteed Guess (CISC → RISC transpilation) accepted at EMNLP 2025; completed MSc (Research) in Computer Vision at MBZUAI.
  • May 2025 Won 1st place at TAMMathon 2025 (AED 25,000).
  • Feb 2025 SALT (parameter-efficient fine-tuning via singular-value adaptation) accepted at BMVC 2025.
  • Nov 2024 Optimizing Brain Tumor Segmentation with MedNeXt accepted at MICCAI 2024 Workshop; 3rd place at BraTS 2024 SSA & Pediatrics Challenge.
  • May 2024 Joined Etisalat as a Machine Learning Engineer, building a RAG chatbot over Excel datasets.
  • Aug 2023 Joined MBZUAI for an MSc by Research in Computer Vision with a full-ride scholarship, under Dr. Mohammad Yaqub.
  • May 2023 Graduated from Jamia Millia Islamia with an outstanding GPA, completing my thesis under Prof. S. M. Muzakkir and Prof. Harish Hirani (co-guide, IIT Delhi).
  • Mar 2022 Started as a Research Intern at IIIT Delhi with Dr. Saket Anand on an autonomous last-mile E-rickshaw (drive-by-wire, LiDAR calibration, YOLOP perception).

Publications

View all →

PERSUADE: A Debate Arena for Benchmarking, Optimizing, and Auditing Persuasion in Language Models

Sarim Hashmi, Samuele Poppi, Salem Lahlou, Martin Takáč, Kshitij Mishra, Nils Lukas

Under review