Huzaifa

PhD Student · Paderborn University

Huzaifa Shaaban Kabakibo

LLM Systems Research·Computer Networks Group

I optimize large language model serving systems: the scheduling, autoscaling, and GPU sharing that maximize utilization and make inference cheaper to run at scale.

Research

I'm a PhD student in the Computer Networks Group at Paderborn University, working on how to run large language model inference more efficiently at the systems level. My current focus is optimizing the serving stack itself: scheduling requests, autoscaling replicas, and multiplexing models across GPUs to squeeze more useful work out of scarce accelerator time. An earlier thread in the same direction looked at cold‑start latency in vLLM: predicting how long an engine takes to spin up so a scheduler can plan around it.

Before starting the PhD I built MoRAGBench, the first benchmarking framework for retrieval‑augmented generation on mobile devices, and worked on running LLM inference efficiently on constrained edge hardware. Across all of it, I'm drawn to the same set of problems: scalable AI systems, systems optimization, and hardware–software co‑design.

Publications

Breaking the Ice: Analyzing Cold Start Latency in vLLM

MLSys 2026

Huzaifa Shaaban Kabakibo, Animesh Trivedi, Lin Wang

Breaks vLLM initialization into fine‑grained phases, shows that cold‑start latency is primarily CPU‑bound rather than GPU‑bound, and builds a predictive model that estimates startup cost to inform autoscaling and scheduling decisions. View at MLSys 2026 →

SelectInfer: Selective Neuron Loading and Computation for On‑Device LLMs

arXiv 2026

Huzaifa Shaaban Kabakibo, Eric Schniedermeyer, Artem Burchanow, Lin Wang

A neuron‑level optimization framework for running LLMs on edge devices. An offline profiler separates task‑specific from general‑purpose neurons, then two techniques cut cost at inference time: selective loading keeps only the most important neurons in memory to shrink the footprint, and selective computation evaluates only the most relevant neurons at runtime, trimming memory and compute while preserving task accuracy. Read on arXiv →

Projects

MoRAGBench

Benchmarking framework for RAG pipelines on mobile devices

  • First benchmarking framework for retrieval‑augmented generation pipelines on Android and mobile hardware.
  • Measures retrieval quality, indexing performance, and end‑to‑end LLM inference across multiple mobile backends.
  • Modular architecture built for reproducible experimentation on constrained devices.
github.com/upb-cn/MoRAGBench →

Experience

Research Assistant — Computer Networks Group

Paderborn University

  • Built a framework to decompose vLLM initialization into fine‑grained steps and analyzed latency across model configurations.
  • Identified cold‑start latency as primarily CPU‑bound and designed a predictive model to guide scheduling.
  • Led the work behind Breaking the Ice, accepted at MLSys 2026.

Research Assistant — Paderborn Centre for Parallel Computing

Paderborn University

  • Automated testing and configuration management with Bash scripting and GitLab CI/CD.
  • Ran measurements and performance analysis on FPGA hardware, with scripts for automated data collection.
  • Troubleshot and debugged hardware/software issues across FPGA‑based workflows.

Full Stack Developer — BEA Technology

Lebanon

  • Developed AI‑driven systems for real‑time operations with a focus on UI/UX.
  • Optimized core systems for data flow and improved database performance, security, and API tokenization.

Education

PhD in Computer Science

Paderborn University · Computer Networks Group — LLM serving systems & inference optimization

M.Sc. Computer Engineering

Paderborn University · Embedded Systems focus · Graduated with distinction (sehr gut, 1.1)

B.Sc. Computer Engineering

Lebanese International University · Distinguished degree, GPA 3.97 / 4.0

Activities & Service

Focus & Languages

Tools & Systems

vLLM CUDA Python C / C++ Bash GitLab CI/CD Linux FAISS Android FPGA Nsight profiling

Languages

Arabic Native
English Fluent
German B1 certified