Georg R. PollakI build agentic AI systems and the infrastructure underneath them.

Five years of high-performance computing — teaching GPU programming at CSCS, benchmarking deep-learning frameworks, pushing Bayesian inference onto GPUs — now applied to LLM infrastructure and agentic systems in private equity.

• 2025 – presentThe platform behind this siteAn LLM gateway with real spend controls, an Avro-typed streaming pipeline, single sign-on, and a host reproducible from one playbook.FastAPILiteLLMKafkaPostgresAnsible
• 2026Reinforcement learning the exit decisionA Q-learning agent that decides when a leveraged buyout should sell, trained live in the browser on a stochastic LBO.TypeScriptQ-learningMonte Carlo
• MSc thesis, ETH ZürichDeep learning for KoraliA deep-learning module for a high-performance scientific computing framework, verified and benchmarked against cuDNN, oneDNN, PyTorch and TensorFlow.C++CUDAcuDNNoneDNN
• Semester project, ETH ZürichGPU-accelerated Bayesian inferenceMoving the dense linear algebra behind approximate Bayesian inference onto the GPU, and measuring where that actually pays off.CUDAR-INLAnvprof