Post-training quantization (NVFP4, FP8, NVIDIA ModelOpt) · self-hosted LLM inference with vLLM and SGLang · AI-native application development. Independent software studio in Bochum, Germany.