Amr Hegazy
MSc Informatics student · Technical University of Munich
Munich, Germany
I'm a first-year MSc Informatics student at the Technical University of Munich. My research is about making large models controllable and efficient. I work on inference-time activation steering for LLM safety, recurrent-depth transformers that recover expressivity without the memory cost of unique layers, and formal safety guarantees for LLM-controlled robots via reachability analysis.
I completed my B.Sc. in Computer Science and Engineering at the German University in Cairo, graduating with highest honors (top 5% of my class), and wrote my bachelor's thesis at TUM Heilbronn on safety-guaranteed use of foundation models in robotics (grade: A+). Before that, I spent two years as a machine learning researcher at Siemens DISW, and I worked as an AI & automations engineer at ResearchPreneurs.
LLM safety & steeringEfficient transformersRecurrent depthSafe roboticsReachability analysis
News
- Started my MSc in Informatics at TUM.
- New preprint: Gated Recurrent Transformers. Code is on GitHub.
- Graduated from the German University in Cairo with highest honors.
- Guiding Giants was accepted at the ACL TrustNLP 2026 workshop.
- Reviewing for ICML 2026 (AI4GOOD & CompLearn workshops) and IROS 2026.
- Paper on ML-based FPGA synthesis-feature prediction published at IEEE ICM 2025.
Publications
See Google Scholar for the latest list.
-
2026
Gated Recurrent Transformers: Expressive Depth through Recurrent Modulation
Preprint, arXiv:2608.15062
A recurrent-depth transformer: a single shared core iterated R times between fixed prelude and coda blocks, with a lightweight GRU-style update gate that keeps per-iteration specialization while reusing weights.
-
2026
Guiding Giants: Lightweight Controllers for Weighted Activation Steering in LLMs
6th Workshop on Trustworthy NLP (TrustNLP @ ACL 2026)
A small trainable controller reads intermediate activations and predicts a global scale plus per-layer weights for a refusal-direction steering patch. It raises refusal rates on harmful prompts without touching the LLM's weights.
-
2026
Meta-heuristic Adaptive Exploration Strategy Through Routing Optimization
ICEENG 2026, International Conference for Innovations in Power and Control
-
2025
Online Data-Driven Reachability Analysis using Zonotopic Recursive Least Squares
Preprint, arXiv:2509.17058
-
2025
Safe LLM-Controlled Robots with Formal Guarantees via Reachability Analysis
Preprint, arXiv:2503.03911
A data-driven reachability-analysis safety layer that formally verifies LLM-issued robot commands without needing an exact system model.
-
2025
AI/ML Based Prediction of FPGA Synthesis Features From RTLs
37th International Conference on Microelectronics (IEEE ICM 2025)
ML models that predict RAM, DSP, and FSM synthesis features directly from RTL, bypassing the traditional synthesis flow. This work came out of my time at Siemens DISW.
-
2025
Keyphrase Trees: Enhancing LLM Prompts with Hierarchical Topic Extraction
3rd Cognitive Models and Artificial Intelligence Conference (AICCONF 2025)
Documents are split into segments and clustered semantically into a hierarchy, which then guides an LLM's keyphrase extraction.
Experience
-
AI & Automations Engineer
ResearchPreneurs · Heilbronn, Germany
- Built an AI-driven recommendation system that matches researchers with companies, cutting manual matching time by 85%.
- Led migration planning from a monolith to a microservices architecture.
- Automated key team workflows, increasing overall productivity 6×.
-
Machine Learning Researcher
Siemens DISW · Cairo, Egypt
- DVT QA: ML models that predict FPGA synthesis features (RAMs, DSPs, FSMs) directly from RTL, published at IEEE ICM 2025.
- Calibre QA: LLMs for regression-test keyword extraction and automatic test-case generation, plus a phenomenological ML model for parasitic capacitance extraction (adjusted R² = 99.6%).
-
E-commerce Intern
Froneri · Cairo, Egypt
- Ran sell-in/sell-out analysis that surfaced market gaps and shaped regional sales strategy. Prototyped a customer-facing e-commerce platform.
-
Software Engineering Intern
Shatablee · Cairo, Egypt
- Built an ML floorplan-recognition platform that detects doors, windows, and zones and exports labeled AutoCAD drafts for 3D modeling, cutting manual drafting effort by 90%.
Education
-
M.Sc. Informatics
Technical University of Munich · Munich
-
Bachelor's Thesis Exchange
Technical University of Munich · Heilbronn
Thesis: Safety-guaranteed utilization of foundation models in robotics via reachability analysis. Grade: A+
-
B.Sc. Computer Science and Engineering
German University in Cairo · Cairo
Excellent with highest honors (GPA 0.80, German scale A+), consistently in the top 5% of my class.
Projects
Gated Recurrent Transformer
2026Official code for the GRT paper: recurrent-depth language models with gated recurrent modulation.
Python · PyTorch
Weighted Activation Steering
2025Implementation of Guiding Giants: controller-driven, per-layer activation steering for LLM safety.
Python · PyTorch · HuggingFace
Database Engine
SystemsA simplified relational database engine with B+ tree indexing, multiple data types, and persistence through serialized pages.
Java
Rahhal Travel Platform
2025Role-based travel platform built on Node.js/TypeScript microservices (auth, booking, payments, notifications) with RabbitMQ events, Stripe, Amadeus, and Google Maps.
React · TypeScript · Docker · MongoDB · RabbitMQ
PintOS
2024Extended the PintOS kernel with user programs, system calls, virtual memory (paging, swapping), and a file system with inodes and directories.
C · x86
ML Circuit Solver
2023Turns a photo of a circuit into a graph and simulates it, using object detection for components and graph reduction.
YOLOv8 · OpenCV · NetworkX
AmrNet
2023A from-scratch deep learning library for building, training, and evaluating neural networks, packaged with a modular API and published on PyPI.
Python · NumPy
Simple Container
SystemsA minimal container runtime in Go, isolating processes with Linux UTS, PID, and mount namespaces.
Go · Linux
More on GitHub, including Haskell, Prolog, and systems projects.
Academic Service
- Reviewer, ICML 2026 (AI4GOOD & CompLearn Workshops)
- Reviewer, IROS 2026
- Reviewer, AAAI 2025 (Main Conference)
Leadership
- Machine Learning Head, QuBit GUC (2023–2024)
- ICPC Trainee, ICPC GUC Community (2022–2026)
- AWS Solutions Architect Associate training (2024)
Skills
- Languages
- Python, C/C++, CUDA, Java, JavaScript/TypeScript, SQL, Go, Haskell, Prolog
- ML
- PyTorch, HuggingFace, vLLM, scikit-learn, OpenCV
- Infra
- Docker, AWS, Kafka, Spark, Airflow, RabbitMQ, Git
- Spoken
- Arabic (native), English (fluent), French (B1), German (A2)