DevelopmentSubsection
AI / LLM
Building features on a component that gives a different answer to the same question twice. Six topics in the order they become relevant: what the model is actually doing, how to ask it properly, how to give it your own data, when to change the model itself, how to run the result in production, and what changes when it can call tools. Testing these features is a discipline of its own and lives in QA.
What is inside
Every page in this group, with what each one covers.
- LLM FundamentalsNext-token prediction, context windows and tokens as the unit of cost — the mental model everything else needs.TokensContext
- Prompt EngineeringThe anatomy of a prompt that works, why examples beat adjectives, and where prompting stops being enough.StructureExamples
- RAG ArchitectureRetrieval as the part that actually decides quality: chunking, hybrid search, reranking and citation checking.RetrievalChunking
- Fine-tuningThe adaptation ladder — prompt, retrieve, then fine-tune — and the honest test for when the last rung is warranted.LoRAWhen not to
- LLMOpsRunning it in production: the eval set as the release gate, the versioned artefact, tracing, and cost control.Eval setsTracing
- AI AgentsThe loop of plan, act, observe — what a tool call really is, and where an agent stops being the right shape.Tool useLoops
Where to start
- First LLM featureFundamentals, then prompt engineering. Most problems people bring to RAG or fine-tuning are prompt problems.
- It needs your own dataRAG architecture — and read the retrieval half carefully, because that is where the quality actually comes from.
- It works and now it shipsLLMOps for running it, then the QA section for testing it — the two are the same problem from two sides.
Elsewhere in the section
- DevelopmentSoftware development is the work of deciding where the boundaries in a system go, what runs on each side of them, and what moving one of those boundaries will cost once the code is live. This section covers 28 topics — architecture, the browser, five backend runtimes, four mobile platforms, language models and the databases underneath all of it. Every page turns on the constraint that actually separates the options, rather than on the API surface.
- AI TestingWhen the same input stops producing the same output, an equality assertion becomes meaningless and a single run stops being evidence. Three topics: what that changes for the person writing the cases, how to judge one answer without spending money to replace a fact with an opinion, and how to tell a real regression from sampling noise. Running an LLM feature in production is LLMOps, in Development — these three own testing it.
- MLOpsEverything that has to be true before a model is allowed near production, and everything that keeps being true afterwards. Three topics: the pipeline that makes a training run reproducible, the serving path where latency and cost become somebody’s problem, and the index that makes similarity search fast enough to use. Running an LLM feature specifically is LLMOps, in Development.
Found this useful?
Share it with someone who is working on the same problem.