Glossary

AI and ML

Baseline: A simple reference model used to compare more advanced models.

Feature Engineering: The process of transforming raw data into useful model inputs.

Overfitting: When a model learns training data patterns too specifically and performs poorly on new data.

Underfitting: When a model is too simple to capture useful patterns.

Calibration: How well predicted probabilities match real-world outcomes.

Embedding: A vector representation of text, image, audio, or other data.

Tokenization: Splitting text into model-readable units.

Context Window: The maximum amount of input and output a model can handle in one request.

Hallucination: An AI output that is unsupported, incorrect, or fabricated.

RAG

RAG: Retrieval-Augmented Generation, a pattern that retrieves external context before generating an answer.

Chunking: Splitting documents into smaller pieces for retrieval.

Vector Search: Searching by semantic similarity between embeddings.

Hybrid Search: Combining semantic search and keyword search.

Reranking: Reordering retrieved results with a more precise scoring model.

Faithfulness: Whether an answer is supported by the provided context.

Citation Correctness: Whether cited sources truly support the answer.

Agents

Tool Calling: Letting a model request structured function/tool execution.

Agent Loop: A repeated cycle of plan, act, observe, and respond.

Human-in-the-Loop: A workflow where a human confirms or reviews important decisions.

Idempotency: Making repeated execution of the same action safe and non-duplicative.

Audit Log: A record of actions, actors, inputs, outputs, and decisions.

Cloud and Operations

IAM: Identity and Access Management.

RBAC: Role-Based Access Control.

Autoscaling: Automatically adjusting compute capacity based on demand.

Observability: Understanding system behavior through logs, metrics, and traces.

RTO: Recovery Time Objective, the maximum acceptable restore time after failure.

RPO: Recovery Point Objective, the maximum acceptable data loss window.

IaC: Infrastructure as Code.

Secret: Sensitive value such as an API key, password, or token.

Product

KPI: Key Performance Indicator.

ROI: Return on Investment.

Pilot: A controlled early deployment used to validate feasibility and value.

Acceptance Criteria: Conditions that must be true for work to be considered complete.

Risk Register: A table of risks, likelihood, impact, mitigation, and owner.