Skip to main content

Large Language Models

15 articles · written by Niraj Kumar

Most engineering writing about LLMs stops at the API call. These articles start after it — at the point where you have to choose between a hosted API and a local model, decide whether fine-tuning is worth it, keep token costs attributable per customer, and prove the model has not regressed since last week.

Topics here include fine-tuning with Unsloth and QLoRA, running local models with Ollama on ordinary developer hardware, evaluating and monitoring fine-tuned models in production, per-tenant token cost tracking with OpenTelemetry, privacy-first architectures for teams that cannot send data to third-party APIs, and lightweight on-device inference.

The recurring theme is that the interesting decisions are operational, not model-selection: what you measure, what you cache, what you keep on your own hardware, and what you are willing to pay per request.

Need this built rather than explained?

Production LangChain & LangGraph systems for RAG, agents, and AI workflows.

LangChain Developer Services

Related topics

← All articles