More about

Nikolai Zhukov, a mathematician, computer scientist, ML Engineer based in Germany since 2011 (german citizen).

Hold diplom in Mathematics and Computer Science (with distinction).

Current research interests - mechanistic interpretability - specifically how transformers represent temporal concepts. My recent work compares PCA-subspace and SAE-latent causal interventions on temporal representations in Gemma 2 2B (post).

Previously :

• deployed Gemma4 locally on Raspberry Pi for ischemic stroke recovery support (Google Gemma 4 hackathon, 2026) post

• built an LLM evaluation framework benchmarking providers across cost, latency, and output quality, integrated with Whisper STT/TTS for plaicube.ai

• applied reinforcement learning to industrial topology optimization at Pragmatic GmbH

• research on GNN-based physics simulation at DFKI Saarbrücken

• data science (4+ years), AI-based document processing system for enterprise clients

• research at DFKI Saarbrücken (Medical AI, Cyber-Physical Systems)

I work mainly on linux with python and pyTorch.