# Systems Engineering: Diagnosing the NVIDIA-SMI Kernel Communication Error on LinuxWhen nvidia-smi fails with "NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver", the NVIDIA Management Library (NVML) is unable to establish communication with the nvidia.ko2 min read·gtzhost.hashnode.dev13h ago
# MLOps Architecture: Deploying LiteLLM Proxy as a Secure Gateway for Ollama and vLLMIn production AI environments, standardizing on a single inference runtime is rarely optimal. Teams frequently pair Ollama for rapid local model iteration with vLLM for high-concurrency continuous bat2 min read·gtzhost.hashnode.dev15h ago
# Systems Engineering: The Economics of Self-Hosted LLM InferenceFor enterprises scaling generative AI, relying on external APIs introduces severe data privacy risks, rate-limit bottlenecks, and unpredictable token-based billing. Commercial LLMs distinguish between2 min read·gtzhost.hashnode.devNov 12
Systems Engineering: Mitigating Storage Bottlenecks (NVMe vs SATA)When architecting bare-metal infrastructure, storage protocol overhead is often overlooked. A common anti-pattern in systems engineering is over-provisioning compute to compensate for application late2 min read·gtzhost.hashnode.devNov 6
Systems Engineering: Provisioning a Secure LLM Inference API with OllamaSystems Engineering: Provisioning a Secure LLM Inference API with Ollama2 min read·gtzhost.hashnode.devOct 24