Training
AI Engineering: LLMs in Production
Go beyond API calls: prompt engineering that survives contact with users, RAG architectures, evaluation, cost control and shipping LLM features that behave in production.
Most AI training stops at "call the API and marvel". This one starts there — because the interesting problems begin after the demo: hallucination control, retrieval quality, latency and cost budgets, evaluation you can trust, and the architecture that keeps an LLM feature maintainable as models change underneath you.
The program is built around a real project: by the end you will have designed, built and evaluated a production-grade AI assistant over your own data, with the same patterns I use in client systems — structured outputs, tool calling, RAG with honest retrieval metrics, and guardrails that fail safely.
Aimed at working developers. You bring a language you're comfortable in (Python, PHP or C# all fine); the concepts transfer.
Curriculum
-
1
LLM fundamentals: tokens, context windows, sampling and what models actually do
-
2
Prompt engineering as engineering: templates, structured outputs, function/tool calling
-
3
RAG deep dive: chunking, embeddings, vector stores, hybrid retrieval, re-ranking
-
4
Evaluation: golden sets, LLM-as-judge honestly, regression testing prompts
-
5
Production concerns: streaming, caching, cost budgets, fallbacks and guardrails
-
6
Agents and multi-step workflows: when they help and when they flail
-
7
Capstone: an assistant over your own data, evaluated and deployed
