I'm a second-year undergraduate in Artificial Intelligence at MBZUAI in Abu Dhabi. I work on agentic and embodied infrastructure: the systems that let AI agents, and the models behind them, run fast and reliably on real hardware, from GPU clusters down to edge devices.
With Prof. Xiaosong Ma, I built MemArena, a benchmark for on-device personal memory assistants, accepted at NeurIPS 2026.
I also build products. I'm co-founder and CTO of MemAI, where we build interactive AI tutoring, and founder of 7D24H, a desktop agent that turns tender documents and research ideas into finished bids and papers.
NeurIPS 2026, Evaluations and Datasets Track
A single-world conversational benchmark generated by the MASim multi-agent simulator: 50 agents over 15 days, 10.3M tokens, and six evaluation dimensions, used to test memory backends and open-weight models on-device.
What the benchmark measures, the three results that stand out, and the full preprint.
中文版Pinning partial GPU allocations per node, fixing multi-node server config, and measuring NUMA effects on MI210.
中文版