Glossary: RAG Terminology and Concepts

Scope: This skill is a quick-reference dictionary — short, scannable definitions with analogies. For deep, multi-layered explorations of how a concept works and why, use /explain.

An interactive glossary that provides clear, multi-layered definitions of RAG terminology. Designed to be a quick reference that also teaches.

Language awareness: See .claude/LANGUAGE_AWARENESS.md.

Step 1: Determine What to Look Up

• If the user provides a term (e.g., /glossary embeddings), define that term.
• If the user provides a module number (e.g., /glossary module 3), list all key terms for that module.
• If no argument is given, present the full glossary organized by category.

Step 2: Present the Definition

For each term, provide a structured definition with these layers:

One-Line Definition

A concise, precise definition in one sentence. Example: "An embedding is a dense numerical vector that represents the semantic meaning of a piece of text."

Analogy

An everyday analogy that makes the concept intuitive. Example: "Think of embeddings like GPS coordinates for meaning — texts with similar meanings have coordinates that are close together on the map of all possible meanings."

Technical Detail

A deeper technical explanation for learners who want to understand the mechanism. Include relevant details like dimensions, algorithms, or mathematical concepts — but keep it accessible.

Code Example

A minimal code snippet in the learner's chosen language that demonstrates the concept in action:

Generate a minimal, runnable example in the learner's language. For instance, if explaining embeddings, show how to create one using the appropriate library.

Common Misconceptions

One or two things people often get wrong about this concept. Example: "Embeddings are not word-for-word encodings — the same word in different contexts will have different embeddings in contextual models."

Related Terms

Links to other glossary terms that are closely related, so the learner can explore connected concepts.

Curriculum Connection

Which module and lesson covers this concept in depth, so the learner can dive deeper if needed.

Step 3: Glossary Categories

Organize terms into these categories for browsing:

Foundations

• RAG (Retrieval-Augmented Generation)
• LLM (Large Language Model)
• Embedding / Vector Embedding
• Token / Tokenization
• Context Window
• Prompt / Prompt Template

Document Processing

• Chunking / Chunk
• Overlap (chunk overlap)
• Document Loader
• Metadata
• Preprocessing / Cleaning

Retrieval

• Vector Database / Vector Store
• Similarity Search / Nearest Neighbor
• Cosine Similarity
• Top-k Retrieval
• BM25 / Sparse Retrieval
• Hybrid Search
• Re-ranking

Generation

• Grounding
• Hallucination
• Faithfulness
• System Prompt
• Few-shot Examples
• Chain-of-Thought

Evaluation

• RAGAS
• Precision / Recall
• MRR (Mean Reciprocal Rank)
• Faithfulness Score
• Answer Relevancy

Advanced Concepts

• Hypothetical Document Embedding (HyDE)
• Multi-hop Retrieval
• Agentic RAG
• Self-RAG
• Query Decomposition
• Recursive Retrieval

Step 4: Interactive Features

• If the learner asks "what does X mean in the context of Y?", tailor the explanation to that specific context.
• If they ask to compare two terms, provide a brief comparison highlighting the key difference.
• If they seem confused, offer the simpler analogy first and build up to the technical detail.

Suggest 2-3 relevant next steps using slash commands:

• /explain — get a deep-dive explanation if you want to go beyond the definition
• /lesson — learn about this concept in a structured lesson with hands-on exercises
• /quiz — test your understanding of the terms and concepts you just reviewed

Guidelines

• Keep definitions precise — do not conflate terms that have distinct meanings
• Use consistent terminology throughout (match what the curriculum uses)
• Update examples to match the learner's tech stack when possible
• Be honest when terms have multiple definitions in the field — note the ambiguity
• This should feel like a knowledgeable colleague explaining things, not a textbook