5 posts

#LLM

A narrower tag about running large language models in production. How to decide between self-hosted models like Llama, Mistral or Qwen and API providers, why RAG over corporate docs is harder than demos suggest, and how to build LLM support that does not frustrate customers. One post tests Hermes 3 as an agent with function calling on your own server.

It overlaps with the AI tag but stays closer to engineering. Start with the open-source vs API comparison, since that answer changes how you build everything else, and read the RAG piece before promising anyone a chatbot that knows your documentation.

Related service: AI assistant · Web application development

← All articles

(01) Internal AI assistants — building them right vs the usual mess Every company in 2026 is building "our internal ChatGPT." Most end up as expensive abandonware within a year. A few become indispensable productivity tools. The difference is in five engineering decisions made at the start. → (02) Choosing between open-source LLMs and API providers in 2026 OpenAI, Anthropic, Google APIs vs self-hosted Llama, Mistral, Qwen. The decision used to be mostly about cost. In 2026 it's about latency, privacy, controllability, compliance, and lock-in. Practical framework for choosing. → (03) RAG over corporate docs — what teams underestimate RAG looks simple in demos: index documents, retrieve chunks, ask LLM. Production RAG over real corporate knowledge is harder than demos suggest. Teams underestimate data quality, chunking strategy, evaluation, and ongoing maintenance. → (04) LLM-powered customer support without making it worse than humans AI customer support is everywhere in 2026, and most of it is worse than the human alternative — slower, evasive, hallucinating, frustrating. A short guide to building LLM support that customers actually prefer over hold music. → (05) Hermes 3 as an agent: function calling and tool use on your own server Hermes 3 from Nous Research is a Llama 3.1 fine-tune tuned for function calling and role-based prompting. What it can do as an agent and why teams pick it over OpenAI when data cannot leave the perimeter. →