Glossary.
Every acronym and term used across AI Engineering material, with the one-line meaning you should be able to say out loud.
| Term | Expansion | What it means |
|---|---|---|
| LLM | Large Language Model | A transformer-based model trained on massive text corpora — the generation engine in Gen AI systems. |
| FM | Foundation Model | A large pre-trained model adapted to many downstream tasks; what managed model-hosting services offer. |
| RAG | Retrieval-Augmented Generation | Grounding generation in retrieved evidence rather than relying only on parametric model knowledge. |
| CRAG | Corrective RAG | A RAG variant where retrieved context is graded for relevance before generation, with rewrite/retry on a failing grade. |
| GraphRAG | Graph-based Retrieval-Augmented Generation | Retrieval over an entity-relationship graph rather than passage similarity alone — valuable for multi-hop relational queries. |
| SFT | Supervised Fine-Tuning | Training on input-output example pairs to teach a model a behavior, format, or domain language. |
| PEFT | Parameter-Efficient Fine-Tuning | A family of methods (LoRA, adapters, prompt tuning) that adapt a model by training a small fraction of its weights. |
| LoRA | Low-Rank Adaptation | Freezes base model weights and learns small low-rank delta matrices — a swappable, cheap-to-train adapter. |
| QLoRA | Quantized LoRA | LoRA applied to a 4-bit quantized base model, enabling fine-tuning of larger models on smaller GPUs. |
| RLHF | Reinforcement Learning from Human Feedback | Trains a reward model from human preference rankings, then optimizes the LLM against it. |
| RLAIF | Reinforcement Learning from AI Feedback | Like RLHF, but the preference signal comes from another model rather than human raters. |
| ANN | Approximate Nearest Neighbor | Vector search that trades a small amount of recall for a large speed gain, instead of scanning every vector exhaustively. HNSW and IVF are common ANN index families. |
| HNSW | Hierarchical Navigable Small World | An approximate nearest-neighbor (ANN) graph index widely used for fast vector similarity search. |
| MRR | Mean Reciprocal Rank | An evaluation metric scoring how high the first correct/relevant result ranked, averaged across queries. |
| LLMOps | LLM Operations | The operational discipline around prompt/model versioning, evaluation gates, monitoring, and rollback for LLM systems. |
| HITL | Human-in-the-Loop | A design pattern requiring explicit human confirmation before a system takes a consequential or irreversible action. |
| MCP | Model Context Protocol | A standardized interface for exposing tools and context to an LLM or agent, independent of any one vendor's API. |
| A2A | Agent-to-Agent | A protocol pattern for structured communication and capability discovery between autonomous agents. |
| PII | Personally Identifiable Information | Data that can identify an individual — a first-class concern for redaction, authorization, and output scanning. |
| PHI | Protected Health Information | Health-related PII subject to additional regulatory controls (e.g. HIPAA) in applicable jurisdictions. |