When, why, how. An honest guide to self-hosting an LLM: cases where it pays off, and cases where it wastes time.
Everyone talks about self-hosting Llama or Mistral. Few actually do it.
Confidentiality, sovereignty, massive constant volume, ultra-low latency, predictable cost at scale.
Llama 3.3 70B, Mistral Large 2, Qwen 2.5 72B, DeepSeek V3.
2x H100 or 1x H200 for quality 70B with 4-bit quantization.
vLLM is the 2026 reference. OpenAI-compatible API out of the box.
From Copilot to Claude, AI assistants radically changed how I code in 2026. Concrete feedback after 18 months of daily production use.
Honest benchmark on 40 real tasks: refactor, debug, API generation, framework migration. Unvarnished verdict between the three generative AI giants.
Released by Anthropic in late 2024, MCP is the protocol that turns LLMs into true agents connected to your stack. Here is why everyone is switching.
I design and ship tailor-made applications. 9 years of experience. Let's talk over a coffee (or a call).
Start a project