Machine Learning Researcher
Building brains
Pushing the boundaries of AI at Siemens R&D. Expertise in developing and deploying enterprise-grade AI solutions—from multimodal foundation models and LLM agents to vision architectures—powered by distributed training pipelines and latency-sensitive inference engines.
Academic & Corporate Experience
Professional Experience
Corporate Experience
Machine Learning R&D
Siemens AG • Berlin
As part of the R&D team at Siemens AG, I specialize in developing deep-learning computer vision models to identify and interpret electrical elements and symbols. My work focuses on leveraging machine learning techniques to construct intelligent, energy-efficient smart grids and enhance predictive maintenance capabilities. I was responsible for building state-of-the-art Computer Vision models and training them on Azure ML studio using Compute Clusters.
Machine Learning - TDI Division
Deutsche Bank • Berlin
Worked in the Technology, Data & Innovation division at Deutsche Bank, focusing on applying machine learning techniques to financial data analysis and process automation. Developed and deployed ML models for various banking applications.
Featured Work
Explore my latest work
A collection of my projects, research papers, and machine learning expertise.
- Machine Learning Projects
Implementations of cutting-edge ML algorithms and research papers, focusing on deep learning and quantum computing applications.
- Research Papers
Public notes on quantum computing and theoretical physics, including mathematical derivations and code implementations.
- Research Notes
Deep dives into AI, physics, mathematics, and computer science. Exploring complex concepts through clear explanations.
Research Papers
Latest Paper Explanations
Deep dives into the most influential research papers in LLMs, computer vision and agentic models.
DeepSeek-R1: A Robust and Responsible Language Model
Various Authors
DeepSeek-R1 introduces novel training techniques and architectural improvements to create a more reliable and controllable language model...
DeepFace: Closing the Gap to Human-Level Performance in Face Verification
Taigman, Yang, Ranzato, Wolf
DeepFace introduces a nine-layer deep neural network architecture that achieves an accuracy of 97.35% on the Labeled Faces in the Wild (LFW) dataset...
Cognitive Architectures for Language Agents
Sumers, Yao, Narasimhan, Griffiths
This paper presents a systematic approach to building language agents with cognitive architectures, exploring the intersection of language models and decision-making systems...
Distillation Scaling Laws
Busbridge, Shidani, Webb, Littwin
This study reveals fundamental patterns in how knowledge distillation effectiveness scales with model size, data quantity, and architectural choices...
BMW Agents: A Framework For Task Automation Through Multi-Agent Collaboration
Crawford, duffy, Evazzade, Foehr
This framework introduces innovative approaches to multi-agent collaboration, enabling complex task automation through distributed intelligence and coordinated decision-making...
Apple Intelligence Foundation Language Models
Various Authors
Apple's approach to developing and deploying foundation models focuses on privacy-preserving techniques and efficient on-device inference...
YOLOv9: Learning Using Programmable Gradient Information
Wang, C., Lyu, S., Zhou, X., et al.
YOLOv9 introduces revolutionary techniques for gradient manipulation during training, enabling more effective feature learning and state-of-the-art detection accuracy...
YOLOv12: Attention-Centric Real-Time Object Detection
Wang, C., Ren, Y., Lyu, S., et al.
YOLOv12 incorporates novel attention mechanisms to enhance feature representation while maintaining the speed advantage of the YOLO architecture...
SAM 2: Segment Anything in Images and Videos
Kirillov, A., Mintun, E., Ravi, N., et al.
SAM 2 introduces significant improvements over its predecessor, including better boundary precision, enhanced zero-shot capabilities, and extended functionality for video segmentation...
Transformer-Squared: Self-Adaptive LLMs
Ranzato, M., Touvron, H., Grave, E., et al.
Transformer-Squared introduces a meta-learning approach for language models, allowing them to reconfigure their parameters on-the-fly for specific tasks without explicit fine-tuning...