Series
Production Large Language Model Engineering
Tokenization, context windows, sampling, and prompting are the four ideas everything else in production LLM work builds on. Here's how they actually work and why they matter once real traffic hits.
- maxwell.kimaiyo
Production Large Language Model Engineering: Part 1
Tokenization, context windows, sampling, and prompting are the four ideas everything else in production LLM work builds on. Here's how they actually work and why they matter once real traffic hits.
AI Engineering9 min readAug 21, 2026
- maxwell.kimaiyo
Production Large Language Model Engineering: Part 2
An LLM application depends on model providers, databases, vector stores, and external APIs — any of which can slow down or fail. Rate limiting, backpressure, retries, timeouts, and circuit breakers are how you keep the system standing anyway.
AI Engineering10 min readAug 21, 2026
- maxwell.kimaiyo
Production Large Language Model Engineering: Part 3
Traditional search matches words. Semantic search matches meaning. Here's how embeddings actually work, why vector search returns irrelevant results more often than people expect, and how to debug it when it does.
AI Engineering6 min readAug 21, 2026
- maxwell.kimaiyo
Production Large Language Model Engineering: Part 4
RAG changes what the model sees, not what it knows. Here's when retrieval-augmented generation is the right tool, what a full RAG pipeline actually involves, and a worked example with country-specific regulations that change monthly.
AI Engineering4 min readAug 21, 2026
- maxwell.kimaiyo
Production Large Language Model Engineering: Part 5
SFT, LoRA, DPO, and RLHF all change a model's behavior — but none of them give it current knowledge. Here's how to decide between fine-tuning and RAG, why they're usually both needed, and what a complete production architecture looks like end to end.
AI Engineering9 min readAug 21, 2026