Key AI Concepts for Spring AI: Models, Prompts, Embeddings, Tokens, Structured Output, and RAG

Before starting hands‑on work with Spring AI, it is important to understand several core AI concepts.

1. AI Model

An AI model is the mathematical structure and algorithm trained on data that enables a computer system to simulate human‑like intelligent behavior, such as recognizing patterns, making predictions, or deciding actions.

Various model types exist for different scenarios:

2. Prompt and Prompt Template

A prompt (or prompt word) is the textual instruction, question, or description that a user provides to an AI model; it acts as the task description for the model.

Writing effective prompts requires clarifying the intent, providing necessary background, specifying constraints, and defining the desired output format and style.

A prompt template is a pre‑designed, structured prompt framework containing placeholders; users fill in the placeholders to generate a complete, customized prompt.

Typical workflow: choose or design a template → fill placeholders with specific information → generate a full prompt → feed it to the AI model → obtain the expected output.

3. Embedding

Embedding converts complex, unstructured data (text, images, audio, video, etc.) into a dense numeric vector that captures semantic meaning, allowing computers to process and compare concepts mathematically.

4. Token

Tokens are the smallest semantic units processed by large language models. In English, one token is roughly 0.75 words; common words occupy a single token, while longer words may be split into multiple tokens, and punctuation counts as separate tokens. In Chinese, each character is typically a token (e.g., "人工智能" = 4 tokens).

5. Structured Output

AI model outputs are usually plain strings, even when formatted as JSON; they are not true JSON objects. Therefore, converting the raw output into a structured format is necessary for downstream programmatic processing.

6. Retrieval‑Augmented Generation (RAG)

RAG (Retrieval‑Augmented Generation) enhances large language model answers by first retrieving up‑to‑date information from external knowledge bases and then feeding that information together with the user query to the model, greatly improving answer accuracy and trustworthiness.

Example: asking a model about the 2025 Nobel Literature laureate. A pure LLM might fabricate an answer because its training data ends in 2023, whereas a RAG‑enabled system would retrieve the latest official data and generate a correct response with source links.

7. Summary

The article covered core AI concepts—models, prompts, embeddings, tokens, structured output, and RAG—that are essential for mastering Spring AI and for a broader understanding of the artificial‑intelligence field.

Future posts will dive deeper into Spring AI with example code, source‑code analysis, and case studies.