Anwar Ibrahim

Machine Learning Research · ML Engineering

Anwar Ibrahim

Junior Research Fellow and PhD Candidate in Artificial Intelligence and Machine Learning

About

I develop reinforcement-learning methods and research software for accelerator beamline control. My doctoral work covers simulation environments, continuous-control training, and comparison with classical optimization methods.

I work across experimental research and production-oriented machine learning, from model development and evaluation to deployment and monitoring. My broader work includes generative models, agentic systems, natural language processing, and computer vision.

01

Experience

Research, engineering, and teaching roles.

Oct 2023–Present

Junior Research Fellow

HSE University · LAMBDA

Develop reinforcement-learning methods and research software for accelerator beamline control, including simulation environments, continuous-control training, benchmarking, and peer-reviewed publication.

Jan 2025–Present

Lecturer

HSE University

Teach Generative Models in Machine Learning and contribute practical sessions to machine-learning education activities.

Dec 2025–Oct 2026

AI Engineer · Agentic Systems

Eolos AI

Developed components for private agentic AI systems built around large language models. Designed multi-step workflows with LangGraph, implemented human-in-the-loop review and feedback paths, managed application data with Convex, and used LangSmith for tracing and evaluation.

Jun 2020–Sep 2023

Machine Learning Engineer

EXPentech

Developed and evaluated machine-learning models, prepared and explored data, and supported model deployment and monitoring.

Anwar Ibrahim with research colleagues inside an accelerator facility
Research visit Accelerator-facility visit with research colleagues.
Anwar Ibrahim giving a presentation about reinforcement learning
Scientific communication Presenting reinforcement-learning work to a research audience.

02

Education

Academic training in artificial intelligence, data science, and software engineering.

Nov 2023–Present

PhD Candidate in Artificial Intelligence and Machine Learning

HSE University

Doctoral research on reinforcement-learning-based control of accelerator beamlines.

2021–Jul 2023

Master’s in Data Science

HSE University

Applied Mathematics and Information Science.

2015–2020

Bachelor’s in Computer Software Engineering

Higher Institute for Applied Sciences and Technology

03

Publications

Peer-reviewed journal articles. Preprints are open-access versions, not separate publications.

RL-ABC system architecture connecting the reinforcement-learning agent, Gymnasium environment, Elegant Wrapper, and simulation engine
Inside RL-ABC The system architecture used to connect an RL agent with the Elegant accelerator simulation. Figure 2 from the open preprint. View source

  1. Computer Physics Communications · Volume 329 · 2026 · Article 110407

    RL-ABC: Reinforcement Learning for Accelerator Beamline Control

    Anwar Ibrahim, Fedor Ratnikov, Maxim Kaledin, Alexey Petrenko, Denis Derkach

    DOI Open preprint Code

  2. International Journal of Modern Physics E · Volume 35, Issue 07 · 2026

    Reinforcement Learning for Accelerator Beamline Control: A Simulation-Based Approach

    Anwar Ibrahim, Alexey Petrenko, Maxim Kaledin, Ehab Suleiman, Fedor Ratnikov, Denis Derkach

    DOI Open preprint

  3. Physics of Particles and Nuclei · Volume 56, Issue 6 · 2025 · pp. 1476–1481

    Optimization of the Accelerator Control by Reinforcement Learning: A Simulation-Based Approach

    Anwar Ibrahim, Denis Derkach, Alexey Petrenko, Fedor Ratnikov, Maxim Kaledin

    DOI Open preprint

04

Conferences

Selected conference presentations and contributions.

Anwar Ibrahim presenting RL-ABC at CHEP 2026
CHEP 2026 · Bangkok Presenting RL-ABC.
Anwar Ibrahim standing beside a diffusion-models poster at CHEP 2026
CHEP 2026 · Poster session At the conference poster session.
Anwar Ibrahim presenting reinforcement-learning research at NUCLEUS 2025
NUCLEUS 2025 Conference presentation.
Anwar Ibrahim at the MMCP 2024 conference in Yerevan
MMCP 2024 · Yerevan Conference participation.
CHEP · Talk

RL-ABC: Reinforcement Learning for Accelerator Beamline Control

View
ACAT · Oral

Optimisation of the Accelerator Control by Reinforcement Learning: A Simulation-Based Approach

View
CERN workshop

Reinforcement Learning for Accelerator Beamline Control

View
NUCLEUS · Talk

Reinforcement Learning for Accelerator Beamline Control

View
MMCP

Optimization of Accelerator Control by Reinforcement Learning

View

05

Selected projects

Public repositories with concrete implementations and documentation.

Research softwarePython

RL-ABC

Framework for converting Elegant accelerator lattice configurations into RL environments, with continuous-control agents and classical optimization baselines.

Repository
Offline reinforcement learningPyTorch

Implicit Q-Learning on Hopper

Implementation of Implicit Q-Learning using D4RL’s Hopper expert dataset, with actor, value, twin-critic, and target networks.

Repository
Explainable reinforcement learningPyTorch

Interpretable RL for Pong

Analysis toolkit for a Pong agent using saliency maps, Q-value inspection, paddle-position analysis, and an interactive dashboard.

Repository
Natural language processingTransformers

Multilingual Contexto

A semantic word-guessing game built with BERT-family models and Streamlit, supporting Arabic, English, and Russian.

Repository
Natural language processingTensorFlow

Disaster Tweet Detector

Transformer-based classification of emergency and disaster-related tweets, with a Streamlit interface for real-time use.

Repository
Computer visionC# / YOLO

Traffic Monitoring System

Vehicle detection, tracking, and speed estimation using YOLO models, a WPF desktop interface, and SQL-based data storage.

Repository

06

Skills & languages

Tools used across research, engineering, and teaching.

Machine learning

  • Reinforcement learning
  • Generative models
  • Diffusion models
  • Natural language processing
  • Computer vision
  • Human-in-the-loop systems

Frameworks & tools

  • Python
  • PyTorch
  • TensorFlow
  • Gymnasium
  • Stable-Baselines3
  • LangChain
  • LangGraph
  • LangSmith
  • Convex
  • Streamlit
  • OpenCV
  • Elegant / SDDS

Additional technologies

  • C#
  • WPF
  • SQL
  • Git
  • Experiment tracking
  • Model deployment

Languages

  • Arabic
  • English
  • Russian

07

Teaching

University instruction and practical machine-learning education.

HSE University

Generative Models in Machine Learning

English-language instruction covering major generative-model families, including variational autoencoders, GANs, autoregressive models, and normalizing flows.

Course page

All-Russian High-Energy Physics School

Practical Machine Learning Track

English-language practical sessions and a reinforcement-learning competition for school participants.

HSE article
Anwar Ibrahim with colleagues at the All-Russian High-Energy Physics School
High-energy physics school At the school programme with colleagues.
Group photograph from the All-Russian High-Energy Physics School
School community Group photograph from the programme.

08

Media coverage

Selected official and national coverage of the accelerator-control research.

HSE University · Science news · 2026

Физики научили ИИ настраивать пучок частиц в ускорителе

The official HSE article explains the RL-ABC study and includes comments from Anwar Ibrahim about the control problem and results.

Read at HSE
Газета.Ru Рамблер/новости Ведомости Наука Mail Naked Science CNews Наука в Сибири 21mm

Additional HSE pages

Practical ML teaching at the High-Energy Physics School PhD pre-defence announcement
Email copied