AI Engineering Glossary¶
Quick definitions with links to handbook modules.
| Term | Definition | Learn more |
|---|---|---|
| Agent | System that uses an LLM to plan, call tools, and act in a loop toward a goal | M11 |
| Agent harness | Runtime wrapping the agent loop: permissions, tools, state, termination, observability | M18 |
| Attention | Mechanism letting models weigh relevance of input tokens | M00 |
| Chunking | Splitting documents into retrieval-sized pieces | M09 |
| Context window | Max tokens a model can process in one request | M01 |
| Embedding | Dense vector representation of text for similarity search | M13 |
| Eval / evaluation | Measuring LLM/agent quality offline or online | M19 |
| Fine-tuning | Training a base model on domain-specific data | M15 |
| Golden dataset | Curated input/output pairs for regression testing | M19 |
| Guardrails | Constraints on model outputs (format, safety, scope) | M14 |
| Hallucination | Model generating plausible but false content | M16 |
| Harness | See Agent harness | M18 |
| LLMOps | Operating LLM apps in production (deploy, monitor, version) | M10 |
| LLM-as-judge | Using an LLM to score another model's outputs | M19 |
| MCP | Model Context Protocol — standard for tool/resource servers | M18 |
| Observability | Traces, logs, metrics for debugging LLM pipelines | M10 |
| Orchestration | Coordinating multiple agents or steps in a workflow | M12 |
| PEFT / LoRA | Parameter-efficient fine-tuning methods | M15 |
| Prompt injection | Attack where user input overrides system instructions | M16 |
| RAG | Retrieval-Augmented Generation — fetch context then generate | M09 |
| ReAct | Reason + Act pattern for tool-using agents | M11 |
| RLHF | Reinforcement Learning from Human Feedback | M07 |
| Tool use | LLM calling external functions/APIs | M11 |
| Trajectory eval | Scoring full agent run (steps, tools, outcome) | M19 |
| Loop engineering | Designing inner/outer agent control cycles | 2026 · Loop Engineering |
| Context engineering | Curating what enters the context window | 2026 · Context Engineering |
| Skills (SKILL.md) | Reusable agent instruction files | 2026 · Skills & Rules |
| Claude Code | Agentic terminal coding harness | 2026 · Claude Code |
| Transformer | Architecture based on self-attention | M06 |
| Vector database | Store optimized for similarity search on embeddings | M13 |