AI app building covers the work of turning an LLM API into a product that real users pay for. The category sits between two adjacent ones: pure prompt engineering (which is shallower) and machine learning (which builds the underlying models). What you're actually doing is integration engineering on top of one or more model APIs, plus the surrounding infrastructure for streaming, retrieval, evals, and feature delivery.
Most AI products in 2026 share the same architecture: a thin client (web or mobile), a backend that orchestrates calls to models and tools, a vector store for retrieval-augmented context, and a queue for long-running jobs. The differentiator is rarely the model — it's the workflow, the data, and the UX.
What you'll work with in these 16 courses
- Model APIs: OpenAI, Anthropic, Gemini, open-source via Replicate or Together
- Streaming responses, structured outputs, JSON-mode, tool use
- RAG: chunking strategies, embeddings, hybrid search, reranking
- Vector stores: pgvector, Pinecone, Weaviate, Qdrant, Turbopuffer
- Evals: golden datasets, LLM-as-judge, regression testing
- Production concerns: rate limits, cost monitoring, prompt-injection defense
