AnyoneLearnAI
Start learning

Reference

Glossary A–Z

Look up any AI term — search or jump by letter. Each entry links to a lesson when one exists.

334 terms

A

  • A/B testingSplitting traffic between **two variants** (prompt, model, retrieval) and comparing metrics to pick
  • Acceptance criteriaDefine “done” with golden tasks, latency bounds, and refusal behavior.
  • Adapter (LoRA)A **small trainable layer** bolted onto a frozen base model — cheaper than full fine-tuning.
  • Adversarial promptA user or document crafted to **trick an AI** into breaking rules — ignoring policies, calling forbi
  • Adverse impactWhen a **neutral-looking hiring practice** disproportionately excludes protected groups — often meas
  • AgentAn **agent** is an AI loop that plans steps, calls tools, reads results, and keeps going until it re
  • Agent handoffPassing state and goals from one agent to another in a multi-step workflow.
  • Agent memoryWhat an **agent loop remembers** across steps — message history, tool results, and optional long-ter
  • Agentic codingUsing an AI **agent** — not just autocomplete — to plan, edit multiple files, run terminal commands,
  • AGINews headlines, research debates, and safety discussions about long-term AI capability — not when de
  • AISoftware that learns patterns from data to make predictions, classifications, or generated content —
  • AI copyrightBefore shipping commercial content, training on third-party corpora, or cloning distinctive styles/v
  • AI disclaimerClear user-facing text that output is **informational, may be wrong**, and is not professional advic
  • AI literacyBaseline skills to use, question, and improve with AI tools safely.
  • AI monitoringAny production LLM or ML endpoint. Pair infra signals (latency, 5xx) with quality signals (eval samp
  • AI regulationLaws, standards, and agency rules that govern **how AI systems are built, deployed, labeled, and aud
  • AI writing briefAn **AI writing brief** specifies audience, purpose, voice, length, must-include facts, and forbidde
  • AlignmentShaping AI systems to **follow human intent and values** — helpful, honest, harmless — not just pred
  • Anthropic APICloud API for **Claude models** — messages API with long context, system prompts, tool use, and visi
  • API keyA secret string that **authenticates your app** to an AI provider's API — like a password for your p
  • API rate limitProvider cap on requests per minute — batch jobs help large embed/index builds stay under limits.
  • Attention mechanismHow models weigh which tokens matter when producing the next token.
  • Audit trailImmutable **log of who changed what** — prompts, models, indexes — for compliance and debugging.
  • AutoGenA Microsoft framework for multi-agent apps built around **conversable agents** — agents that talk to
  • Availability targetThe **desired percentage of successful requests** over a window — feeds SLOs and SLAs.

B

  • Backlog prioritizationChoose the smallest experiment that validates an outcome before big agent builds.
  • Backup snapshotA **point-in-time copy** of indexes, configs, or databases used to restore after corruption or delet
  • Baseline metricA **reference number from stable production** — last week’s p95 latency or citation rate — used to d
  • Batch inferenceRunning many model calls in one scheduled batch instead of one sync request at a time.
  • Batch jobAn async job that processes many API requests offline — poll until complete.
  • Batching (inference)Running **multiple model requests together** in one forward pass to improve GPU throughput — often a
  • BenchmarkA **standardized test suite** used to compare models on tasks like math, coding, safety, or retrieva
  • BiasSystematic **unfair skew** in AI outputs — often reflecting biased training data or design choices t
  • Blameless postmortemAn incident review focused on **systems and process fixes**, not punishing individuals for mistakes.
  • Blast radiusHow much damage one failure or leak can cause — users affected, data exposed, budget burned.
  • Blue-green deploymentRun two identical environments — switch traffic atomically after validation.
  • BM25Classic keyword ranking function — strong baseline paired with vector search in hybrid RAG.
  • Burn rateBurn rate measures how fast you are consuming an error budget — useful for alerting before an SLO is

C

  • Calculus and gradientsA **gradient** is the direction and steepness of a loss function's slope — training nudges model wei
  • Canary deploymentReleasing a change to **a small slice of traffic first**, watching metrics, then promoting or rollin
  • Capacity planningEstimating **compute, tokens, and concurrency** needed to meet latency and cost targets at expected
  • Chain-of-thought promptingWord problems, planning, and tasks where intermediate checks help you audit the path. Prefer structu
  • ChatbotAn app that **accepts user messages and returns AI replies** — often with memory, tools, or retrieve
  • ChromaA **lightweight, in-process vector database** — it runs inside your Python app (no separate server r
  • Chunk overlapRepeated text between adjacent chunks so sentences split across boundaries still retrieve.
  • ChunkingSplitting long documents into **smaller pieces** before embedding or retrieval so RAG systems can fi
  • Circuit breakerPattern that stops calling a failing dependency after repeated errors — prevents retry storms.
  • Citation (in AI answers)Pointing to **where a claim came from** — a doc title, URL, chunk ID, or page — so readers can verif
  • ClassificationPredicting a **discrete label** — spam/not, yes/no, cat/dog — from inputs or scores.
  • Claude CodeA terminal-based coding agent from Anthropic — you describe a task in plain language, and it reads t
  • ClusteringExploratory analysis: segment users, theme open-ended survey answers, or bucket documents before lab
  • Coding assistantA **coding assistant** suggests or edits code inside a chat or IDE. It speeds drafting; you still ow
  • CompletionThe model's **generated continuation** of a prompt — the next tokens it predicts, whether a single p
  • Compliance reviewCross-functional check that a product or workflow meets **policies, regulations, and internal contro
  • Computer visionAI that **interprets images and video** — detect objects, read text (OCR), segment regions, estimate
  • Confusion matrixA table of **true vs predicted labels** — shows false positives and false negatives, not just overal
  • ConsentPermission before using someone's likeness, voice, or data with AI.
  • Constitutional AITrain models with written principles — critique and revise harmful outputs.
  • Content moderationFiltering or blocking harmful user/model content before display or storage.
  • Context windowThe maximum amount of text (measured in **tokens**) a model can read and remember **in one conversat
  • Conversation memoryTechniques to keep multi-turn chats useful within **token limits** — trim, summarize, or store facts
  • Convolutional neural networkVision networks that scan images with learned filters — core of many image models.
  • Cosine similarityA score from **-1 to 1** measuring how aligned two vectors are — common for comparing embeddings.
  • Cost allocationTagging **token spend by team, feature, or customer** so you know who consumed what.
  • Cost per tokenHow providers **price API usage** — separate rates for input tokens (your prompt) and output tokens
  • CrewAIA framework for assembling a **"crew" of role-based agents** (e.g. researcher, writer, editor) that
  • Cross-encoderA model that **scores query–document pairs together** — slower than bi-encoders but sharper for re-r
  • Cross-validationSplitting data into **multiple train/test folds** so you estimate performance more reliably than a s
  • CursorAn AI-first code editor (a fork of VS Code) with a built-in **Agent mode** that can read your repo,
  • Customer communicationTimely, accurate **messages to users** during outages — what broke, what you are doing, what they sh

D

  • Dark launchShip new behavior to production **hidden from most users** — often via feature flags at 0–5% traffic
  • Data driftWhen **live inputs change** from what you trained or indexed on — new slang, products, or user behav
  • Data governanceRules for who can access, store, and train on organizational data.
  • Data lineageWhere training and RAG data came from — who can access and when it was refreshed.
  • Data pipelineAutomated flow from raw data → clean tables → features → model or index.
  • Dataset splitDividing examples into **train / validation / test** so you measure generalization, not memorization
  • Dead letter queueStorage for **failed automation jobs** after retries exhaust — so nothing silently disappears.
  • Decision treeA **decision tree** splits data with if/else rules to predict a class or value. Easy to visualize; e
  • Deep learningPerception and generation problems with lots of examples: images, speech, language, recommendation r
  • DeepEvalAn open-source LLM evaluation framework that plugs into your existing test runner (like pytest) — yo
  • DeepfakeAI-generated or altered **video, audio, or images** that make a real person appear to say or do some
  • Dependency checkVerifying upstream services — embedding API, vector DB, secrets — before accepting user traffic.
  • Diffusion modelA generative model that **starts from noise and iteratively denoises** into an image (or other signa
  • Direct Preference Optimization (DPO)A simpler alternative to full RLHF for aligning a model to human preferences — instead of training a
  • Disaster recovery (DR)Plans and runbooks to **restore AI services** after region outages, data corruption, or dependency f
  • DistillationTraining a **smaller "student" model** to mimic a larger "teacher" model's outputs — keeping much of
  • DSPyA framework that treats prompting as **programming, not string templating** — you declare the input/

E

  • EmbeddingAn **embedding** is a list of numbers that captures the meaning of text (or other media) so computer
  • Embedding batchGrouping many texts into **one embeddings API job** (or batch endpoint) instead of one call per row.
  • Embedding indexStored vectors for chunks — built once, queried at retrieval time.
  • Embedding modelA model that turns text into **fixed-size vectors** for similarity search — not for chat completions
  • Error budgetAn error budget is the allowed unreliability in an SLO period — the room you have for incidents and
  • Escalation policyWritten rules for **when and how** to pull in leadership, legal, comms, or vendor support during inc
  • Eval gateA **CI or deploy checkpoint** that blocks release when golden tasks score below threshold.
  • Eval setA fixed list of **input → expected-quality** tasks used to compare prompts, models, or releases.
  • Experiment trackingLogging **prompt, model, data, and metric** for each training or eval run so you can compare and rep
  • Exponential backoffRetry strategy: **wait longer after each failure** (1s, 2s, 4s…) to avoid hammering a rate-limited A

F

  • FailoverRegion outages, failed readiness probes, and planned maintenance with minimal downtime.
  • FairnessFair treatment across groups in AI outcomes — not identical outputs for everyone.
  • FaithfulnessHow closely a model output **matches the provided context** — no extra claims beyond what sources su
  • Fallback modelBackup model when primary fails rate limits or errors — often smaller/cheaper.
  • Fault injectionDeliberately **breaking dependencies** in controlled tests to verify retries, fallbacks, and runbook
  • Feature engineeringChoosing and transforming **input columns** so a model can learn — scaling, encoding categories, or
  • Feature flagToggle features per user cohort without redeploying — pairs with experiments.
  • Few-shot promptingGiving the model **a few input→output examples** in the prompt so it copies the pattern on new input
  • Fiduciary dutyLegal obligation to act **in a client’s best interest** — applies to advisors, trustees, and some co
  • Financial disclosurePublic or investor-facing **statements about financial condition** — subject to audit, regulation, a
  • Fine-tuningConsistent brand voice, specialized classification, or repeated output shapes where prompt engineeri
  • Function callingWhen an LLM **returns a structured request** to run a function (API, database, calculator) instead o

G

  • Game dayA planned **failure injection or drill** in staging (or controlled prod) to test runbooks and on-cal
  • Generative AIAI that **creates new content** — text, images, audio, video — from a prompt or input, rather than o
  • GGUFFile format for packaged quantized local LLM weights — common in Ollama and llama.cpp.
  • GitAny project you might revisit, share, or deploy. Use commits as checkpoints before and after AI-assi
  • GitHubTeam code review, CI checks, open-source contribution, and storing project history remotely.
  • GitHub CopilotAn AI pair-programmer plugin for popular editors (VS Code, JetBrains, etc.) that started as inline a
  • Golden signalsThe four golden signals are latency, traffic, errors, and saturation — a starting dashboard for any
  • Golden taskA **fixed test input** with known quality criteria — the backbone of eval gates and regression tests
  • GPUGraphics Processing Unit — hardware that **parallelizes matrix math**, making neural network inferen
  • Graceful degradationServing **reduced but useful** responses when dependencies fail — instead of total outage.
  • Gradient descentAn optimization loop that **nudges model weights** in the direction that reduces loss — the core “le
  • GrafanaGrafana visualizes metrics and logs so teams can see latency, errors, and custom AI quality signals
  • GroundednessWhether an answer **sticks to retrieved or cited sources** instead of inventing facts.
  • GroundingTying an answer to **specific source text or data** so the model is less likely to invent facts.
  • Guardrail metricA **guardrail metric** triggers pause or rollback when AI quality, safety, or cost crosses a red lin
  • GuardrailsRules, filters, and checks that **limit what an AI system can say or do** — content policies, tool a

H

  • HallucinationWhen an AI states something **false but sounds confident** — invented citations, wrong dates, fake p
  • Health checkA health check endpoint tells load balancers whether a model server is ready to take traffic.
  • Hiring biasSystematic **unfair disadvantage** to candidates based on protected attributes or proxy signals in s
  • HistogramA histogram metric buckets observed values (like latency) so you can estimate percentiles such as p5
  • Hugging FaceAn open platform for **models, datasets, and inference** — hub hosting, Transformers library, and sp
  • Hugging Face TransformersAn open-source Python library that gives you a single, consistent way to **load, run, and fine-tune*
  • Human approvalRequiring a **qualified person to confirm** an action before it affects customers, employees, or reg
  • Human in the loopHumans approve high-stakes AI outputs before they take effect.
  • Hybrid searchCombining **keyword search** (exact matches, BM25) with **semantic/vector search** so you catch both

I

  • IdempotencyProcessing the **same event twice** without duplicate side effects — critical for webhook retries.
  • Incident commanderThe **single coordinator** during a major incident — owns timeline, tasks, and comms while specialis
  • Incident responseStructured steps when **production AI fails** — contain harm, rollback, investigate, communicate.
  • Incident severityA **graded scale** (often Sev-1 to Sev-4) describing user impact and response urgency.
  • InferenceEvery production call: chat completions, RAG retrieval embeddings, classification, TTS. Optimize inf
  • Inference endpointHTTP (or gRPC) URL where a deployed model accepts inputs and returns predictions.
  • Input filterServer-side checks on **user content before it reaches the model** — pattern blocks, length limits,

J

  • JailbreakA prompt or trick that **bypasses an AI's safety rules** — getting it to ignore policies, reveal sec
  • JSON modeAPI setting that nudges the model to reply with parseable JSON instead of prose.
  • JSON schemaA contract describing allowed JSON keys and types — models and validators use it for machine-readabl

K

  • Kill switchAn instant **off** control for risky behavior — disables a feature flag or tool without redeploying
  • KV cacheExplaining why long contexts cost more and why caching prefixes helps latency.

L

  • LangChainA popular **Python/JS framework** for chaining LLM calls — prompts, retrievers, tools, agents, and m
  • LangfuseAn open-source LLM observability platform — self-hostable or managed — that traces calls, tracks cos
  • LangGraphA framework (from the LangChain team) for building agents as an explicit **graph of steps** — nodes
  • LangSmithA hosted tracing and evaluation platform (from the LangChain team) that logs every LLM call in your
  • LatencyHow long users wait for an AI response — from sending a prompt to seeing the first token (TTFT) or t
  • Launch readinessGo/no-go checklist for AI betas and GA.
  • Legal holdA directive to **preserve records** when litigation or investigation is likely — overrides normal de
  • Legal reviewQualified **human or counsel examination** of contracts, policies, and regulatory correspondence bef
  • Linear algebra for MLWhen you need to understand why embeddings are vectors, how matrix multiplication powers neural nets
  • Liveness probeA lightweight check that asks **“Is the process alive?”** — if it fails, orchestrators restart the c
  • LlamaIndexA framework focused on **connecting LLMs to your data** — loaders for files/APIs/databases, chunking
  • LLMA **Large Language Model** predicts the next piece of text from patterns learned during training — u
  • LLM as judgeUsing one model to score another's outputs against rubrics — fast but needs human calibration.
  • LLM tracingRecording **prompts, tool calls, latencies, and outputs** per request so you can debug production AI
  • Load testingSimulating **realistic or peak traffic** against an API to measure latency, errors, and resource lim
  • Local inferenceServing a model on your machine or VPC instead of a public cloud API.
  • Local LLMA **local LLM** runs on your machine or private hardware instead of a vendor chat API. Useful for pr
  • LogprobsLog probabilities per token — useful for confidence hints and debugging.
  • LoRACustom tone, domain style, or task format when you have example data but not a huge GPU budget.
  • Loss functionA **loss function** scores how wrong a model’s prediction is during training. Optimizers push parame
  • Loss functions (overview)A **loss function** scores how wrong a model's predictions are — training minimizes this score so th

M

  • Machine LearningA branch of AI where programs **improve from examples** instead of following only hand-written rules
  • Material informationFacts that a **reasonable investor would consider important** when making decisions — must be handle
  • matplotlibPython’s common **plotting library** — scatter plots, histograms, and line charts to see patterns be
  • Max tokensAn API setting that caps **how long the model's reply** can be — measured in tokens, not words.
  • MCPBuilding agents that need shared access to GitHub, Slack, local files, or internal APIs without cust
  • Meeting notes AIAfter meetings with clear agendas; when you can paste notes or a transcript you are allowed to proce
  • Metadata filter (vector search)Restrict similarity search to chunks matching **tags** — tenant, date, doc type — before ranking vec
  • Metric cardinalityCardinality is how many unique time series a metric creates; high-cardinality labels (user IDs) can
  • Mixture of experts (MoE)A model architecture built from many smaller "expert" sub-networks, where a router picks only a hand
  • Model cardA standard doc summarizing a model's limits, data, and intended use.
  • Model compressionShrink models via quantization or pruning for faster/cheaper inference.
  • Model router / gatewayA layer in front of your app's LLM calls that picks **which model** handles each request — by task t
  • Model routingSending each request to the right model size or provider for cost and quality.
  • Model servingAnytime users or other services call a model continuously. Serving choices dominate cost and latency
  • Model versioningTrack which weights/prompt snapshot serves each environment.
  • Model weightsThe **learned numbers** inside a neural network that encode patterns from training — billions of par
  • Multi-agent systemA **multi-agent system** coordinates several LLM agents (researcher, critic, coder) that pass messag
  • MultimodalDocument Q&A over screenshots, image captioning, voice + text agents, video understanding, or any jo

N

  • Narrow AIAI that excels at **one task or domain** — spam filtering, chess, speech-to-text — without general h
  • Neural networkLayers of weighted connections that learn patterns from data.
  • NLPSearch, translation, summarization, sentiment analysis, chatbots, and entity extraction from documen
  • North star metricThe **north star metric** is the single primary outcome that proves your AI feature creates user or
  • NumPyA Python library for **fast arrays of numbers** — the foundation pandas, scikit-learn, and PyTorch a

O

  • Object-oriented programmingOrganizing code with classes and objects — reusable state and methods.
  • ObservabilityMeasuring **logs, metrics, and traces** from live AI systems so you detect drift, cost spikes, and q
  • Offer letter reviewHR and legal **sign-off on compensation, equity, and terms** before an employment offer is sent — ne
  • OllamaA popular tool to **download and run open-weight LLMs locally** with a simple CLI and API.
  • On-callThe **rotation that responds** to production alerts outside business hours.
  • On-device AIModels running on your hardware — data stays local, useful for privacy and offline.
  • Open weightsModel **parameters published for download** so you can run or fine-tune on your own hardware — licen
  • Open-source modelPublicly released model weights you can run locally or fine-tune — check license terms.
  • OpenAI APICloud HTTP API for **ChatGPT-class models** — chat completions, embeddings, images, audio, and assis
  • Ops checklistA repeatable list of **monitoring, security, and recovery items** to verify before ship or on-call h
  • Orchestrator agentAn agent that **plans, delegates, and merges** work from specialist worker agents instead of running
  • OTLPOTLP (OpenTelemetry Protocol) is a standard way to export traces, metrics, and logs from apps to col
  • Output validationChecking model responses against **schema, regex, or policy rules** before showing or storing them.
  • OverfittingWhen a model **memorizes training examples** but fails on new data — great accuracy in the lab, poor

P

  • P95 latencyThe response time below which **95% of requests** finish — catches slow tail better than average alo
  • pandasA Python library for **tables of data** — load CSVs, filter rows, compute stats, and prep features f
  • PEFTCustom style, format, or domain phrasing when full fine-tune is too expensive.
  • Phased rolloutA **phased rollout** ships AI features to a small cohort first — canary, beta, GA — with metrics at
  • PII redactionStrip names, SSNs, and secrets before logging or sending to models.
  • PineconeA **managed, serverless vector database** — you send it embeddings and metadata, it handles indexing
  • PostmortemA **blameless write-up** after an incident — timeline, root cause, and follow-up actions.
  • Pre-trainingThe expensive first stage where an LLM reads massive text to learn general language patterns — befor
  • Precision@kFraction of **top-k retrieved chunks that are relevant** — measures noise in the context window.
  • Presentation AIFirst-pass structure for internal updates and teaching decks when you already know the story.
  • PrivacyKeeping personal data out of models and logs unless policy allows.
  • Probability for MLClassification with uncertain outcomes, calibrating spam filters, choosing decision thresholds, and
  • Probe endpointAn HTTP route (or exec check) platforms poll for **liveness or readiness** — kept fast and idempoten
  • Product requirementsPRDs for AI features should name eval gates, human review points, and metrics — not just “add a chat
  • Production AI architectureMoving from prototype to users: auth, logging, retrieval indexes, guardrails, canaries, and on-call
  • Production readinessEvidence that a system meets **minimum ops standards** before taking real traffic — monitoring, roll
  • PrometheusPrometheus scrapes and stores time-series metrics — request rates, latencies, and custom counters yo
  • PromptThe instructions and context you give an AI model — your question, task, examples, and constraints.
  • Prompt cachingReusing **cached prefix tokens** (system prompt, long docs) so repeat calls pay less for the same co
  • Prompt contractA **prompt contract** turns a vague ask into a testable job: audience, deliverable, sources of truth
  • Prompt engineeringThe practice of **designing instructions, examples, and constraints** so models produce reliable, us
  • Prompt injectionAssistants that read email, browse the web, or retrieve documents can be tricked into leaking secret
  • Prompt templateReusable prompt skeleton with slots for user data — keeps format consistent across requests.
  • promptfooA CLI/config-driven tool for **testing and comparing prompts** — you write a YAML file of test cases
  • PyTorchAn open-source deep learning framework — the most widely used one for research and increasingly prod

Q

  • QdrantAn **open-source vector database** written in Rust, known for fast filtered search and built-in vect
  • Quality gateAn automated **check that blocks release** when quality metrics fail — golden queries, citation chec
  • QuantizationCompressing model weights to **fewer bits** (e.g. 16-bit → 8-bit or 4-bit) so models run faster and

R

  • RAG- Answers can cite sources you control (policies, product docs, tickets).
  • RAGASAn open-source evaluation library built specifically for **RAG pipelines** — it scores answers on me
  • Random forestA **random forest** averages many decision trees trained on bootstrap samples and random feature sub
  • Rate limitA cap on **how many API requests or tokens** you can use per minute — providers enforce it to keep s
  • Re-rankingA second pass that **re-scores retrieved chunks** with a smarter (often slower) model so the best pa
  • Readiness probeChecks **“Can this instance safely take traffic?”** — fails when dependencies (vector DB, model gate
  • Reasoning modelAn LLM configuration that spends **extra compute on stepwise thinking** before answering — better on
  • Recall@kWhether the **correct chunk appears in the top k** retrieved results — retrieval quality before gene
  • Red teamingDeliberately **attacking your own AI system** to find failures — unsafe outputs, leaks, bad tool cal
  • Regulated adviceGuidance in domains where **licensed professionals** or strict rules apply — legal, medical, tax, in
  • Reinforcement learningTraining by **trial and reward** — an agent takes actions, gets feedback, and learns policies that m
  • Release freezeA temporary halt on **risky deploys and flag rollouts** while error budget burns or incidents are op
  • Response formatAPI field controlling output shape — JSON schema, JSON mode, or free text.
  • Responsible AIBuilding and using AI with safety, fairness, and accountability in mind.
  • RetrievalThe step that **finds relevant documents or chunks** before the model generates an answer — the "R"
  • Reward modelScorer that ranks outputs during RLHF — proxy for human preferences.
  • Risk registerA **risk register** lists AI failure modes — hallucination, PII leak, tool abuse — with owners and m
  • Risk tierClassification of AI use by potential harm — stricter rules for higher tiers.
  • RLHFAligning chat models after supervised fine-tuning when you can collect preference data and run rewar
  • RollbackReverting a **live AI deployment** to the last known-good model, prompt, or retrieval index when qua
  • Rollback planDocumented steps to revert model, prompt, or infra when metrics breach SLOs.
  • Root cause analysisFinding the **underlying failure** — bad index, prompt, model, or infra — not just the symptom.
  • Routing policyRules that pick fast vs reasoning vs local models per request.
  • RPO (recovery point objective)The **maximum acceptable data loss** measured in time — how far back you can rewind after failure.
  • RTO (recovery time objective)The **maximum acceptable downtime** before service must be restored after an incident.
  • RunbookA **step-by-step playbook** for common incidents — who to page, what to rollback, where logs live.

S

  • Safety filterLayer that blocks policy-violating prompts or outputs — often before the main model runs.
  • Schema validationChecking parsed JSON against required keys/types before your app acts on it.
  • scikit-learnA Python library for **classical machine learning** — decision trees, random forests, clustering, re
  • Scope creepName it in reviews: “Auto-send refunds is creep — defer to v2 with evals.”
  • Secrets managerCentral store for API keys, DB passwords, and signing secrets — injected at runtime, not committed t
  • Secrets rotationRegularly **replacing credentials** — API keys, DB passwords, signing secrets — without service down
  • Semantic cacheReturning a **stored answer** when a new question is semantically near a past query — skip LLM call
  • Semantic searchSearch by **meaning**, not just exact keywords — powered by embeddings and similarity.
  • Service recoveryRestoring **normal operation** after an incident — rollback, hotfix, or traffic shift back to health
  • Sev-1 (severity 1)The **highest urgency incident** — major user-facing outage or active data breach requiring immediat
  • Shadow deploymentRun new version alongside prod without user traffic — validate before cutover.
  • Similarity searchFinding items whose **meaning** is close to a query — usually by comparing embedding vectors with di
  • SLA (service level agreement)A **contractual promise** to customers — often with credits if you miss uptime or latency targets.
  • SLO (service level objective)A **target for reliability or quality** — e.g. p95 latency under 2s or citation rate above 85% — tha
  • Small language model (SLM)A language model built to run with far fewer parameters (roughly under ~10B) than flagship models —
  • Smoke testA smoke test is a tiny post-deploy check that the happy path still works before you send real traffi
  • Span (tracing)One **timed unit of work** in a trace — e.g. embed query, retrieve chunks, call chat completion.
  • Speculative decodingDraft model predicts tokens; target model verifies — cuts latency for some workloads.
  • Speech recognitionMeeting transcripts, captions, voice commands, call-center notes, and any workflow where spoken word
  • Spreadsheet AIRepetitive formula construction, regex/cleanup helpers, and explaining cryptic errors — with a hand-
  • SSE (Server-Sent Events)An HTTP pattern where the **server pushes chunks** to the browser over one long response — common fo
  • Statistics for MLEvaluating classifiers, comparing A/B model tests, checking class imbalance, reporting precision/rec
  • Status pageA public or customer-facing page showing **current service health** and incident updates.
  • Stop sequenceSpecial tokens or strings that tell the model to end generation early.
  • Streaming (LLM responses)Sending model output **token-by-token** (or chunk-by-chunk) over HTTP instead of waiting for the ful
  • Stress testPushing a system **beyond expected load** until it errors — finding breaking points and failure mode
  • Structured outputForcing the model to reply in a **machine-readable shape** — JSON, XML, or a fixed schema — instead
  • Success metricsBefore beta: define north-star (value) and guardrail (risk) metrics with thresholds.
  • Supervised learningMachine learning from **labeled examples** — each input comes with the correct answer the model shou
  • Support AI playbookA **support AI playbook** defines when an assistant may draft replies, which knowledge it may cite,
  • Sync APIRequest/response API for interactive latency — user waits for one result.
  • Synthetic dataBootstrapping eval sets or formatting examples when real data is scarce.
  • Synthetic mediaContent **created or heavily modified by AI** — images, video, audio, text — especially when it coul
  • Synthetic monitoringAutomated **fake user requests** run on a schedule to detect outages before customers report them.
  • System messageThe API role that sets **global instructions** for the assistant — tone, rules, and scope for the wh
  • System promptHidden instructions that set the AI's **role, rules, and tone** for a whole session — before the use

T

  • Tabletop exerciseA **walk-through drill** where the team talks through an incident scenario without touching producti
  • TemperatureA dial that controls how **random** vs **predictable** an LLM's word choices are. Low = safer and re
  • TensorFlowGoogle's open-source deep learning framework, often used through its high-level **Keras** API — defi
  • Text-to-imageAI that **generates pictures from text prompts** — diffusion and related models map words to pixels.
  • ThresholdA **cutoff score** that turns a continuous model output into a yes/no (or class) decision.
  • TokenA **token** is a small chunk of text — often a word, part of a word, or punctuation — that the model
  • Token budgetA **cap on tokens** per request, user, or day — prevents runaway spend from long contexts or agent l
  • Token quotaA **token quota** caps how many model tokens a user, tenant, or job may consume in a period.
  • TokenizerThe piece of an LLM stack that **splits text into tokens** — the units models read, price, and count
  • Tool (LLM tool use)A **tool** is an external action an LLM can request — search, calculator, database query, email send
  • Tool allowlistAn explicit list of **tool names the model may call** — everything else is rejected at runtime.
  • Tool schemaJSON description of a callable tool's name, parameters, and types.
  • Top-kKeeping only the k highest-scoring retrieval hits before re-rank or generation.
  • Top-p (nucleus sampling)A sampling setting that limits the model to the **smallest set of likely next tokens** whose combine
  • Traffic drainGradually **stop sending new requests** to unhealthy or retiring instances before shutdown or deploy
  • Traffic splitSending only a **fraction of requests** to a new version while the rest stay on stable — core of can
  • TransformerA **transformer** is a neural network architecture that uses **attention** to weigh how much each to
  • TransparencyClear disclosure when content or decisions involve AI.
  • TTS (text-to-speech)Narration, accessibility, product voiceovers, IVR prototypes, and learning how voice direction (pace

U

  • Unsupervised learningFinding **patterns in data without labels** — grouping similar items, detecting anomalies, or reduci
  • UptimeThe fraction of time a service is **available and serving successful responses** — often rolled up m
  • User messageThe API role for **what the human (or app) asks** in this turn — the main task input.
  • User storyA **user story** states who benefits, what they need, and why — in one sentence — before scoping AI

V

  • VectorA **vector** is an ordered list of numbers. In AI, embeddings are vectors that place meaning in a sp
  • Vector databaseA database optimized to **store embeddings and run similarity search** fast — Pinecone, pgvector, Ch
  • Vector indexData structure (HNSW, IVF, etc.) for fast nearest-neighbor search over embeddings.
  • Verification barA **verification bar** is the minimum evidence you require before trusting an AI output — calculatio
  • Vibe codingBuilding software mostly by describing what you want in natural language and letting an AI agent wri
  • Virtual environmentA **virtual environment** isolates Python packages for one project so dependencies do not collide wi
  • Vision encoderThe part of a **multimodal model** that turns images into tokens the LLM can reason over.
  • vLLMHigh-throughput LLM serving library — batching and KV cache optimized inference.
  • Voice cloningYour own consented voice for demos, accessibility, or product narration — with disclosure when liste

W

  • Warm standbyA **pre-provisioned backup environment** kept ready (but not serving traffic) for fast failover.
  • WeaviateAn **open-source vector database** that can run self-hosted or managed, with built-in hybrid search
  • WebhookAn **HTTP callback** when an event happens — new row, form submit, payment — triggering your automat
  • Webhook signatureA webhook signature lets you verify that an incoming callback really came from the provider and was
  • Workflow automationChaining **triggers, AI steps, and app actions** (email, CRM, Sheets) with human checkpoints.
  • Workflow orchestrationCoordinating **multi-step automations** — triggers, LLM calls, branches, retries — as one durable fl
  • Workflow triggerThe **event that starts an automation** — form submit, new email, schedule, webhook.

Z

  • Zero-shot promptingAsking the model to do a task **with instructions only** — no example pairs in the prompt.
← Reference hub