Skip to content

98 - Glossary

Every key term from the Introductions of all guides, A to Z, with a link to the guide that explains it. Generated by tools/build_glossary.py from the "Key terms" tables; do not edit by hand.

Some terms appear in several guides with a meaning specific to that context (for example "Token" in HTTP authentication and in LLMs).

Jump to: A B C D E F G H I J K L M N O P Q R S T U V W Z

A

Term Meaning Guide
Action Reusable building block from the Marketplace (actions/checkout@v4) 44 - GitHub Actions (CI/CD)
Activate Switch the terminal to use the environment's Python 11 - Python Virtual Environment
Adapter The small trained LoRA weights 37 - Fine-tuning
Agent LLM + tools + loop that pursues a goal autonomously 32 - AI Agents
Agentic RAG The LLM decides when and what to search, using a search tool 31 - RAG (Retrieval-Augmented Generation)
Alias Short name for a command (ls -> Get-ChildItem) 03 - Terminal and PowerShell
Anchor Link target created from a heading (#1-headings) 02 - Markdown
ANN Approximate nearest neighbour: fast, slightly inexact search 30 - Embeddings and Vector Databases
Answers file .copier-answers.yml in the generated project: template source, version and answers 51 - Project Templates
API Application programming interface: the agreed way one program asks another to do something 01 - Core Concepts: Code, Packages, APIs and SDKs
API Interface that lets programs talk to each other 40 - FastAPI
API gateway The single entry point that routes requests to services (here: the Nginx proxy) 50 - Project Structure: Python Microservices + Frontend
API key A secret string that proves who you are to an API (and who pays) 01 - Core Concepts: Code, Packages, APIs and SDKs
API key Secret that identifies and bills you; keep in env vars 27 - LLM APIs
App string module:attribute Where Uvicorn finds the app, e.g. app.main:app 41 - Uvicorn (ASGI Server)
Arrow In-memory columnar format shared by Polars, DuckDB, pandas 19 - Polars and DuckDB
Artifact Files saved from a run (reports, builds) 44 - GitHub Actions (CI/CD)
Artist Anything drawn: lines, text, patches 21 - Matplotlib
ASGI Async interface between Python web servers and apps 41 - Uvicorn (ASGI Server)
ASGI app Object the server calls (your app = FastAPI()) 41 - Uvicorn (ASGI Server)
ASGI server (Uvicorn) The program that runs your app and handles connections 40 - FastAPI
Assertion A check that must be true 15 - pytest
await Pause here until the awaited thing finishes; let other tasks run 14 - Async Python
Axes One plot area inside a figure (with its x and y axis) 21 - Matplotlib
Axes-level function Draws one plot into a Matplotlib Axes (boxplot) 22 - Seaborn
Axis A dimension: axis=0 rows direction, axis=1 columns direction 17 - NumPy

B

Term Meaning Guide
Backend Where state is stored (local file, Azure Storage, Terraform Cloud) 47 - Terraform
Backpropagation Computing gradients from the loss back through the layers (loss.backward()) 24 - PyTorch
Base model Model before your fine-tuning (use an instruct model for chat tasks) 37 - Fine-tuning
Base vs instruct model Base continues text; instruct / chat follows instructions 25 - Hugging Face
Bash The default shell on most Linux systems 04 - Linux
Batch API Async bulk processing at lower cost 27 - LLM APIs
Batch / epoch Group of samples per step / one full pass over the data 24 - PyTorch
Blocking call A normal function that waits without letting others run (time.sleep, requests.get) 14 - Async Python
Body / payload The data sent, usually JSON 09 - HTTP and APIs
Branch An independent line of development 05 - Git and GitHub
Breakpoint A line where the debugger pauses 06 - VS Code
Broadcasting Automatic stretching of arrays with different shapes 17 - NumPy
Broker The queue storage that holds jobs (Redis, RabbitMQ) 42 - Redis, Caching and Task Queues
Buffering Proxy collects the response before sending (breaks streaming if on) 45 - Nginx, Reverse Proxy and HTTPS
Build context The folder Docker sends to the builder; files outside it cannot be copied into the image 50 - Project Structure: Python Microservices + Frontend

C

Term Meaning Guide
Cache hit / miss Found in cache / not found (compute and store) 42 - Redis, Caching and Task Queues
Caching Keep expensive objects (models, clients, indexes) between reruns 39 - AI User Interfaces (Streamlit, Gradio, Chainlit)
Catastrophic forgetting Loses general abilities after narrow training 37 - Fine-tuning
Cell One block of code or Markdown 16 - Jupyter
Chain Fixed sequence of steps (prompt -> model -> parser) 33 - Agent Frameworks
Chain of thought Letting the model reason before the final answer 28 - Prompt Engineering
Character class A set of allowed characters, e.g. [a-z] or \d 07 - Regex
Checkpointer Saves state so runs can resume / be inspected 33 - Agent Frameworks
Checkpoint / model ID organisation/model-name, e.g. sentence-transformers/all-MiniLM-L6-v2 25 - Hugging Face
Chunk A piece of a document stored and retrieved as a unit 31 - RAG (Retrieval-Augmented Generation)
Chunk overlap Repeated text between neighbouring chunks so ideas are not cut in half 31 - RAG (Retrieval-Augmented Generation)
Citation Reference to the source of each claim 31 - RAG (Retrieval-Augmented Generation)
CLI Command-line interface: a program you use by typing commands and flags (git, uv, docker) 01 - Core Concepts: Code, Packages, APIs and SDKs
Client The object (or program) that sends requests; in SDKs, client = anthropic.Anthropic() 01 - Core Concepts: Code, Packages, APIs and SDKs
Client Connection inside the host to one server 34 - MCP (Model Context Protocol)
Client / server Who asks / who answers 09 - HTTP and APIs
Client tools Run by your code 29 - Tool Use (Function Calling)
Cluster Control plane + worker nodes 46 - Kubernetes
Cmdlet A PowerShell command (Get-ChildItem) 03 - Terminal and PowerShell
Code-based grading Deterministic checks (exact match, regex, JSON valid, tests pass) 35 - Evals and Observability
Code fence Three backticks that start / end a code block 02 - Markdown
Coercion Converting compatible values ("1" -> 1) 13 - Pydantic
Collection / index A named set of vectors in a vector DB 30 - Embeddings and Vector Databases
Colormap (cmap) Mapping from values to colours 21 - Matplotlib
Command Palette Ctrl+Shift+P: search box for every command 06 - VS Code
Commit A saved snapshot with a message 05 - Git and GitHub
Completion / output Tokens the model generates 26 - LLM Fundamentals
Compose Tool to run several containers from one YAML file 43 - Docker
Compose override file A second Compose file whose settings are merged over the first (for example for development) 50 - Project Structure: Python Microservices + Frontend
Computer use Agent operates a GUI (screenshots, clicks, typing) 32 - AI Agents
Concurrency Many tasks in progress at once (interleaved) 14 - Async Python
ConfigMap / Secret Configuration / sensitive values injected into pods 46 - Kubernetes
Conflict Two branches changed the same lines; you decide the result 05 - Git and GitHub
Container A packaged, isolated copy of an app with everything it needs to run (Docker) 01 - Core Concepts: Code, Packages, APIs and SDKs
Container Running instance of an image 43 - Docker
Content block One piece of the response: text, thinking, tool_use, ... 27 - LLM APIs
Context engineering Curating all information in the context window 28 - Prompt Engineering
Context length (num_ctx) Max tokens per request the server allocates for 36 - Local and Self-Hosted LLMs
Context window Max tokens the model can consider at once (prompt + output) 26 - LLM Fundamentals
Continued pretraining More next-token training on raw domain text 37 - Fine-tuning
Cookiecutter Older, widely used Python templating tool; generates once, no built-in update 51 - Project Templates
Copier Python tool that renders templates from questions and can update generated projects later 51 - Project Templates
copier.yml Copier's configuration: questions, defaults, validators and settings 51 - Project Templates
Coroutine What calling an async def function returns; it does nothing until awaited 14 - Async Python
Corpus / knowledge base All documents you can retrieve from 31 - RAG (Retrieval-Augmented Generation)
Cosine similarity Similarity of direction between two vectors (-1 to 1; higher = more similar) 30 - Embeddings and Vector Databases
Coverage Share of code lines executed by tests 15 - pytest
Crew Group of role-based agents with tasks (CrewAI) 33 - Agent Frameworks

D

Term Meaning Guide
Database A collection of tables managed by a database system 20 - SQL
Data exfiltration Sneaking private data out (e.g. via a URL or email) 38 - AI Security and Responsible AI
DataFrame A table: rows and named columns 18 - Pandas
Data leakage Information from the test data or the answer sneaks into training 23 - Scikit-learn
Data ownership Rule that only one service reads and writes a given database 50 - Project Structure: Python Microservices + Frontend
Data source Read-only lookup of something that already exists 47 - Terraform
DDL / DML Statements that define structure (CREATE) / change data (INSERT) 20 - SQL
Deallocate Stop the VM and its compute billing (disk still billed) 49 - Azure VM + Linux + Ollama
Delimiters / XML tags Markers that separate instructions from data (<document>...</document>) 28 - Prompt Engineering
Dependency A package your project needs in order to run 01 - Core Concepts: Code, Packages, APIs and SDKs
Dependency A package your project needs (pandas, fastapi) 12 - uv
Dependency group Extra packages only for some uses, e.g. dev (pytest, ruff) 12 - uv
Dependency injection FastAPI passes shared things (DB, settings) into endpoints via Depends 40 - FastAPI
Deployment Keeps N identical pods running; handles rolling updates 46 - Kubernetes
Deserialize / load / parse Text -> Python object (json.loads) 08 - YAML, JSON, TOML and .env
Device Where computation runs: cpu, cuda (NVIDIA GPU), mps (Apple GPU) 24 - PyTorch
Dimension Length of the vector (384, 768, 1024, 1536, 3072 ...) 30 - Embeddings and Vector Databases
Direct injection The user themselves types the malicious instructions ("jailbreak") 38 - AI Security and Responsible AI
Distance Opposite of similarity (smaller = closer) 30 - Embeddings and Vector Databases
Distillation Training a small model on outputs of a big model 37 - Fine-tuning
Distribution (distro) A packaged Linux version: Ubuntu, Debian, Fedora 04 - Linux
DNS A record Maps a domain name to your server's IP address 45 - Nginx, Reverse Proxy and HTTPS
Dockerfile Recipe that builds an image 43 - Docker
DPI Resolution of saved images 21 - Matplotlib
Drift Real infrastructure changed outside Terraform 47 - Terraform
dtype The single data type of all elements (int64, float64) 17 - NumPy
dtype Data type of a column (int64, float64, object, datetime64) 18 - Pandas

E

Term Meaning Guide
Embedding model Model that converts text to vectors 30 - Embeddings and Vector Databases
Embedding model Model that turns text into vectors for search / RAG 49 - Azure VM + Linux + Ollama
Embedding / vector List of floats representing meaning 30 - Embeddings and Vector Databases
Endpoint One URL + method of a web API, e.g. POST /v1/messages 01 - Core Concepts: Code, Packages, APIs and SDKs
Endpoint A URL + method that does one thing (POST /v1/messages) 09 - HTTP and APIs
Endpoint / route A URL + method your API answers (GET /items) 40 - FastAPI
Environment Deployment target with its own secrets and protection rules 44 - GitHub Actions (CI/CD)
Environment variable A named setting given to a program by the operating system (ANTHROPIC_API_KEY) 01 - Core Concepts: Code, Packages, APIs and SDKs
Epoch One pass over the training data 37 - Fine-tuning
Eval set / dataset Collection of test cases (input + expected / reference + metadata) 35 - Evals and Observability
Event loop The scheduler that runs coroutines and switches between them 14 - Async Python
Event loop asyncio scheduler handling many connections in one process (14) 41 - Uvicorn (ASGI Server)
Event / trigger (on) What starts the workflow 44 - GitHub Actions (CI/CD)
Exception Python's way of signalling an error; it stops the program unless your code handles it 01 - Core Concepts: Code, Packages, APIs and SDKs
Excessive agency Model has more tools / permissions / autonomy than needed 38 - AI Security and Responsible AI
Execution count The number [5] showing the order cells ran 16 - Jupyter
Expression Polars building block: pl.col("a") * 2 19 - Polars and DuckDB
Extension Add-on that adds features 06 - VS Code

F

Term Meaning Guide
Facet One small plot per category value (col=, row=) 22 - Seaborn
Features (X) Input columns the model learns from 23 - Scikit-learn
Few-shot / multishot Instructions plus a few input -> output examples 28 - Prompt Engineering
Field One attribute with a type (and optional rules / default) 13 - Pydantic
Figure The whole image / window 21 - Matplotlib
Figure-level function Creates a whole figure, can split into facets (catplot) 22 - Seaborn
Fine-tuning / post-training Further training to follow instructions, be helpful and safe 26 - LLM Fundamentals
Fixture Reusable setup (data, client, temp folder) injected by name 15 - pytest
Flag Option that changes matching, e.g. ignore case 07 - Regex
Flaky test Sometimes passes, sometimes fails (timing, randomness, network) 15 - pytest
Foreign key Column that links to another table's primary key 20 - SQL
Forward pass Input through the model to get predictions 24 - PyTorch
Framework A larger structure that calls your code at the right moments (FastAPI, React, pytest) 01 - Core Concepts: Code, Packages, APIs and SDKs
Full fine-tuning Update all weights (needs lots of GPU memory) 37 - Fine-tuning

G

Term Meaning Guide
Gated model Needs you to accept a licence and log in (e.g. some Llama models) 25 - Hugging Face
GFM GitHub Flavored Markdown: standard Markdown plus tables, task lists, alerts 02 - Markdown
GGUF File format for quantized models used by llama.cpp and Ollama 36 - Local and Self-Hosted LLMs
GitHub template repository A repository marked as a template so "Use this template" creates a copy with fresh history 51 - Project Templates
Graceful shutdown Finish in-flight requests before exiting 41 - Uvicorn (ASGI Server)
Grader / scorer Function that scores an output 35 - Evals and Observability
Gradient Direction and size to change each weight to reduce loss 24 - PyTorch
Graph Nodes (steps) + edges (transitions) + shared state (LangGraph) 33 - Agent Frameworks
Grounding Basing answers on provided sources / tools 26 - LLM Fundamentals
Grounding / faithfulness Answer supported by the retrieved text 31 - RAG (Retrieval-Augmented Generation)
Group Part of a pattern in ( ) you can extract separately 07 - Regex
groupby Split rows into groups, then aggregate each 18 - Pandas
Guardrail Check that runs on input / output and can block it 33 - Agent Frameworks
Guardrails Limits and checks on inputs, actions and outputs 32 - AI Agents
Guardrails Input / output checks and policies around the model 38 - AI Security and Responsible AI

H

Term Meaning Guide
Hallucination Confident but false output 26 - LLM Fundamentals
Handoff One agent passes control to another (OpenAI Agents SDK) 33 - Agent Frameworks
Harness The code that runs the loop, executes tools, manages context and limits 32 - AI Agents
HCL The configuration language of Terraform 47 - Terraform
Header Key-value metadata (auth, content type) 09 - HTTP and APIs
Helm Package manager for Kubernetes apps (charts) 46 - Kubernetes
HNSW Popular graph-based ANN index 30 - Embeddings and Vector Databases
Host AI app using MCP (Claude Code, Claude Desktop, IDE, your app) 34 - MCP (Model Context Protocol)
HPA Horizontal Pod Autoscaler: scales replicas on metrics 46 - Kubernetes
HTTP method The action: GET read, POST create, PUT/PATCH update, DELETE remove 40 - FastAPI
hue Column used to colour groups 22 - Seaborn
Human-in-the-loop (HITL) A person approves or corrects steps 32 - AI Agents
Hybrid search Combining vector similarity with keyword (BM25) search 30 - Embeddings and Vector Databases
Hybrid search Vector + keyword search combined 31 - RAG (Retrieval-Augmented Generation)
Hyperparameter Setting you choose before training (max_depth) 23 - Scikit-learn

I

Term Meaning Guide
Idempotent Repeating the request has the same effect as doing it once (GET, PUT, DELETE) 09 - HTTP and APIs
Idempotent Running a job twice has the same effect as once (safe retries) 42 - Redis, Caching and Task Queues
Image Read-only package / template of an app 43 - Docker
Indentation Leading spaces that define structure in YAML 08 - YAML, JSON, TOML and .env
Indentation Spaces at the start of a line that define code blocks 10 - Python Basics
Index Row labels (default 0, 1, 2, ...) 18 - Pandas
Index Structure that makes lookups on a column faster 20 - SQL
Indirect injection Malicious instructions hidden in content the model reads (web page, email, document, tool result) 38 - AI Security and Responsible AI
Inference Using a trained model to predict (no training) 24 - PyTorch
Inference Running the model to get output 26 - LLM Fundamentals
Inference server Program serving a model over HTTP (Ollama, vLLM, llama-server, TGI) 36 - Local and Self-Hosted LLMs
Ingress HTTP(S) routing from outside to Services 46 - Kubernetes
Instruction tuning SFT with instruction-following examples 37 - Fine-tuning
IntelliSense Autocomplete and code understanding 06 - VS Code
Interpreter A program that reads source code and runs it line by line (python is the Python interpreter) 01 - Core Concepts: Code, Packages, APIs and SDKs
Interpreter The Python executable VS Code uses to run your code (pick your .venv) 06 - VS Code
Interpreter The program that runs Python code (python) 10 - Python Basics
Interpreter The python executable that runs your code 11 - Python Virtual Environment

J

Term Meaning Guide
Jailbreak Tricking a model into ignoring its safety rules 38 - AI Security and Responsible AI
Jinja The templating language behind {{ variable }}, {% if %} and filters like upper 51 - Project Templates
.jinja suffix Marks files Copier must render; the suffix is removed in the output 51 - Project Templates
Job A set of steps that runs on one runner; jobs run in parallel unless linked with needs 44 - GitHub Actions (CI/CD)
Job / task One unit of background work 42 - Redis, Caching and Task Queues
JOIN Combine rows from tables by a key 20 - SQL
JSON The text format most web APIs use for data ({"name": "Ada", "age": 36}) 01 - Core Concepts: Code, Packages, APIs and SDKs
JSON Text format for data sent and received 40 - FastAPI
JSON-RPC The message format MCP uses 34 - MCP (Model Context Protocol)
JSON Schema Machine-readable description of the model, generated automatically 13 - Pydantic

K

Term Meaning Guide
KDE Smooth estimate of a distribution 22 - Seaborn
Keep-alive Reusing one TCP connection for several requests 41 - Uvicorn (ASGI Server)
Kernel The Python process that runs the cells 16 - Jupyter
Key / value Name and its data in a mapping 08 - YAML, JSON, TOML and .env
Key / value Name and stored data (user:42:name -> "Ana") 42 - Redis, Caching and Task Queues
Knowledge cutoff Date after which the model has no training data 26 - LLM Fundamentals
kubectl The command-line tool to talk to the cluster 46 - Kubernetes
KV cache Memory storing attention data for the current context; grows with context length 36 - Local and Self-Hosted LLMs

L

Term Meaning Guide
Label / selector Key-value tags / how objects find each other 46 - Kubernetes
Latency Time until the (first) response 26 - LLM Fundamentals
Latency, TTFT Total response time / time to first token 35 - Evals and Observability
Layer One cached step of an image build 43 - Docker
Layered architecture Splitting code into routes (HTTP), services (business logic) and repositories (data access) 50 - Project Structure: Python Microservices + Frontend
Lazy frame A query plan that runs only on .collect() 19 - Polars and DuckDB
Learning rate Size of each update step 24 - PyTorch
Least privilege Give only the minimum access needed 38 - AI Security and Responsible AI
Let's Encrypt / certbot Free certificate authority / tool to get and renew certificates 45 - Nginx, Reverse Proxy and HTTPS
Library Reusable code you call from your program to do one job (HTTP, dates, math) 01 - Core Concepts: Code, Packages, APIs and SDKs
Library A package made by others (pandas, requests) 10 - Python Basics
Lifespan Startup / shutdown events of the app (load models, open pools) 41 - Uvicorn (ASGI Server)
LLM Large language model that generates text 49 - Azure VM + Linux + Ollama
LLM-as-judge A model grades outputs using a rubric 35 - Evals and Observability
Local Named expression to avoid repetition 47 - Terraform
Localhost "This computer"; http://localhost:8000 is a server running on your own machine 01 - Core Concepts: Code, Packages, APIs and SDKs
Location Rules for a URL path inside a server block 45 - Nginx, Reverse Proxy and HTTPS
Lock file A file recording the exact version of every installed package (uv.lock), so installs are repeatable 01 - Core Concepts: Code, Packages, APIs and SDKs
Logits Raw output scores before softmax 25 - Hugging Face
LoRA / QLoRA Low-rank adapters / LoRA on a 4-bit quantized base model 37 - Fine-tuning
Loss Number measuring how wrong the predictions are 24 - PyTorch

M

Term Meaning Guide
Magic command Jupyter helper starting with % or %% (%timeit) 16 - Jupyter
Managed identity An identity Azure gives your app so it can access other resources without passwords 48 - Azure
Match A piece of text the pattern fits 07 - Regex
Matrix Run the same job for several versions / OSes 44 - GitHub Actions (CI/CD)
max_tokens Hard cap on output tokens for this response 27 - LLM APIs
MCP Inspector Developer tool to test servers interactively 34 - MCP (Model Context Protocol)
Memory (short / long term) Current context window / stored knowledge across sessions 32 - AI Agents
Merge Combine the changes of two branches 05 - Git and GitHub
Metacharacter A symbol with special meaning (. * + ? ^ $) 07 - Regex
Metadata / payload Extra fields stored with each vector (source, date, user) for filtering 30 - Embeddings and Vector Databases
Method (verb) What to do: GET, POST, PUT, PATCH, DELETE 09 - HTTP and APIs
Microservice A small, separately deployable service that owns one business capability and its data 50 - Project Structure: Python Microservices + Frontend
Mock / fake / stub A stand-in for a real dependency (API, database, LLM) 15 - pytest
Model A class inheriting BaseModel that defines fields and types 13 - Pydantic
Model card README of a model: purpose, licence, limitations 25 - Hugging Face
Model / estimator An algorithm object with fit and predict 23 - Scikit-learn
Model Hub Website hosting models, datasets and Spaces 25 - Hugging Face
Model ID Exact model string, e.g. claude-opus-5 27 - LLM APIs
Model / module A network: layers + a forward method (nn.Module) 24 - PyTorch
Modular monolith One deployable app split internally into well separated modules; often the better first step 50 - Project Structure: Python Microservices + Frontend
Module One Python file you can import 01 - Core Concepts: Code, Packages, APIs and SDKs
Module Reusable group of resources (like a function) 47 - Terraform
Module / package A .py file / a folder of modules you can import 10 - Python Basics
Monkeypatch Temporarily replace an attribute or env var during a test 15 - pytest
Monorepo One Git repository containing several services and apps 50 - Project Structure: Python Microservices + Frontend
Multi-agent Several agents with roles (orchestrator, workers, reviewers) 32 - AI Agents
Multimodal Can take images / PDFs / audio as input 26 - LLM Fundamentals

N

Term Meaning Guide
Namespace Folder-like isolation for objects 46 - Kubernetes
NaN Missing value 18 - Pandas
ndarray NumPy's N-dimensional array 17 - NumPy
Nearest neighbours (kNN) The k most similar vectors 30 - Embeddings and Vector Databases
Nesting Objects / lists inside other objects / lists 08 - YAML, JSON, TOML and .env
Node A machine (VM) that runs pods 46 - Kubernetes
Normalisation Scaling vectors to length 1 (then dot product = cosine) 30 - Embeddings and Vector Databases
Notebook The .ipynb document with cells 16 - Jupyter
NSG Network Security Group: firewall rules for the VM 49 - Azure VM + Linux + Ollama

O

Term Meaning Guide
Offline / online eval On a fixed dataset before release / on live traffic 35 - Evals and Observability
Offloading Keeping some layers on CPU RAM when VRAM is too small (slower) 36 - Local and Self-Hosted LLMs
OIDC Keyless login from GitHub to a cloud (no stored cloud passwords) 44 - GitHub Actions (CI/CD)
OLAP Analytical queries over many rows (DuckDB) vs OLTP transactions (Postgres) 19 - Polars and DuckDB
OpenAI-compatible API Same endpoints as OpenAI (/v1/chat/completions), so OpenAI clients work 36 - Local and Self-Hosted LLMs
Open-weight model Model whose weights can be downloaded 36 - Local and Self-Hosted LLMs
Optimizer Algorithm that updates weights (SGD, Adam, AdamW) 24 - PyTorch
Overfitting Model memorises training data and fails on new data 23 - Scikit-learn
Overfitting Great on training data, poor on new data 24 - PyTorch
Overfitting Memorises training examples, worse on new inputs 37 - Fine-tuning

P

Term Meaning Guide
Package A bundle of modules with a name and version, installed with a package manager (requests, anthropic) 01 - Core Concepts: Code, Packages, APIs and SDKs
Package index / registry The online store of packages (PyPI for Python, npm for JavaScript) 01 - Core Concepts: Code, Packages, APIs and SDKs
Package manager A tool that downloads and installs packages and their dependencies (uv, pip, npm) 01 - Core Concepts: Code, Packages, APIs and SDKs
Package manager Installs software from trusted repositories (apt on Ubuntu) 04 - Linux
Pairwise comparison Judge picks the better of two outputs 35 - Evals and Observability
Palette Set of colours 22 - Seaborn
Parallelism Many tasks running at the exact same time on several cores 14 - Async Python
Parallel tool calls Several tool_use blocks in one response 29 - Tool Use (Function Calling)
Parameters (7B, 70B) Number of weights, in billions; bigger = smarter but heavier 36 - Local and Self-Hosted LLMs
Parameters / weights The learnable numbers inside the model 24 - PyTorch
Parameters / weights Learned numbers in the model (billions) 26 - LLM Fundamentals
Parametrize Run the same test with many inputs 15 - pytest
Parquet Compressed columnar file format (the standard for analytics) 19 - Polars and DuckDB
Pass@k Share of tasks solved in at least one of k attempts 35 - Evals and Observability
Path Location of a file or folder (D:\Projects\app) 03 - Terminal and PowerShell
PATH (env var) List of folders where the shell looks for programs 03 - Terminal and PowerShell
Pattern The regex itself, e.g. \d+ 07 - Regex
PEFT Parameter-efficient fine-tuning (train a small number of extra weights) 37 - Fine-tuning
Permissions Who may read, write or execute each file 04 - Linux
PII Personally identifiable information 38 - AI Security and Responsible AI
pip Package installer for Python 11 - Python Virtual Environment
Pipeline Passing output to the next command with \| 03 - Terminal and PowerShell
Pipeline Preprocessing steps + model as one object 23 - Scikit-learn
Pipeline One-line helper for a task 25 - Hugging Face
pip / uv Tools to install packages 10 - Python Basics
Placeholder A {{ ... }} expression replaced by an answer when the template is rendered 51 - Project Templates
Plan / apply / destroy Preview / execute / remove changes 47 - Terraform
Planning Making a step list before acting 32 - AI Agents
Pod Smallest unit: one (or a few tightly coupled) containers with shared network 46 - Kubernetes
Port 80 / 443 HTTP / HTTPS 45 - Nginx, Reverse Proxy and HTTPS
Port mapping Connect a host port to a container port (-p 8000:8000) 43 - Docker
Predicate pushdown Filtering while reading, so less data is loaded 19 - Polars and DuckDB
Preference tuning (DPO, ORPO, RLHF) Training on "better vs worse" answer pairs 37 - Fine-tuning
Pretraining Learning to predict text on a huge corpus 26 - LLM Fundamentals
Primary key Unique ID of each row 20 - SQL
Probe Health check (liveness, readiness, startup) 46 - Kubernetes
Producer Code that enqueues jobs (your API) 42 - Redis, Caching and Task Queues
Projection pushdown Reading only the needed columns 19 - Polars and DuckDB
Project template A reusable starting point for new projects or parts of projects 51 - Project Templates
Prompt Everything you send: system instructions, messages, documents 26 - LLM Fundamentals
Prompt Reusable template (user-controlled, often shown as slash commands) 34 - MCP (Model Context Protocol)
Prompt caching Provider reuses a repeated prompt prefix: cheaper and faster 27 - LLM APIs
Prompt chaining Splitting a task into several prompts, output of one feeds the next 28 - Prompt Engineering
Prompt injection Untrusted text trying to override your instructions (38) 28 - Prompt Engineering
Prompt injection Input that overrides or hijacks the developer's instructions 38 - AI Security and Responsible AI
Prompt template Prompt with placeholders filled at runtime 28 - Prompt Engineering
Provider Plugin that talks to a platform's API (azurerm, aws, kubernetes) 47 - Terraform
Proxy headers X-Forwarded-For / X-Forwarded-Proto set by a reverse proxy 41 - Uvicorn (ASGI Server)
Pub/Sub Publish messages to channels; subscribers receive them live 42 - Redis, Caching and Task Queues
Pull request (PR) A GitHub request to merge a branch, with review 05 - Git and GitHub
Push / pull Upload / download commits 05 - Git and GitHub
Pydantic model Class that defines and validates data shape 40 - FastAPI
PyPI The Python Package Index where libraries are published 10 - Python Basics
pyplot (plt) The quick, state-based interface 21 - Matplotlib
pyproject.toml Project file: name, Python version, dependencies, tool settings 12 - uv
.python-version Which Python version this project uses 12 - uv

Q

Term Meaning Guide
Q4_K_M, Q5_K_M, Q8_0 Common GGUF quantization levels (bits + method) 36 - Local and Self-Hosted LLMs
Quantifier How many times something repeats (+, {3}) 07 - Regex
Quantization Storing weights with fewer bits (8, 5, 4 ...) to save memory 36 - Local and Self-Hosted LLMs
Query A SELECT statement that reads data 20 - SQL
Query string ?key=value parameters in the URL 09 - HTTP and APIs

R

Term Meaning Guide
Rate limit Max requests allowed per time window 09 - HTTP and APIs
Rate limit Max requests / tokens per minute for your account 27 - LLM APIs
RBAC Role-based access control: who can do what, on which scope 48 - Azure
ReAct Pattern: Reason -> Act -> Observe, repeated 32 - AI Agents
Reasoning / thinking Model spends tokens thinking before answering 26 - LLM Fundamentals
Red teaming Deliberately attacking your own system to find weaknesses 38 - AI Security and Responsible AI
Reference answer Known good answer to compare against 35 - Evals and Observability
Reflection Agent reviews its own output and improves it 32 - AI Agents
Region Physical data-centre location (swedencentral, westeurope, northeurope) 48 - Azure
Registry Storage for images (Docker Hub, Azure Container Registry) 43 - Docker
Regression Something that worked before and now fails 35 - Evals and Observability
Regression test Test that reproduces a fixed bug so it never returns 15 - pytest
Reload Restart automatically on code changes (development only) 41 - Uvicorn (ASGI Server)
Remote / origin The copy of the repo on a server (GitHub) 05 - Git and GitHub
Render Turning Markdown symbols into formatted output 02 - Markdown
ReplicaSet Created by a Deployment to maintain the pod count 46 - Kubernetes
Repository pattern A class that hides database queries behind simple methods like add and get 50 - Project Structure: Python Microservices + Frontend
Repository (repo) A project folder tracked by Git (history lives in .git/) 05 - Git and GitHub
Request ID An ID attached to a request and passed between services so logs can be joined 50 - Project Structure: Python Microservices + Frontend
Request / response The message you send to an API and the message it sends back 01 - Core Concepts: Code, Packages, APIs and SDKs
requirements.txt Text file listing the project's packages and versions 11 - Python Virtual Environment
Reranker Model that re-scores retrieved chunks more precisely 31 - RAG (Retrieval-Augmented Generation)
Rerun (Streamlit) Script executes again after each user interaction 39 - AI User Interfaces (Streamlit, Gradio, Chainlit)
Resolve Work out a set of versions that are compatible with each other 12 - uv
Resource Readable data identified by a URI (app-controlled) 34 - MCP (Model Context Protocol)
Resource One infrastructure object to manage (azurerm_storage_account) 47 - Terraform
Resource One service instance: VM, storage account, web app ... 48 - Azure
Resource group Folder in Azure that holds related resources 49 - Azure VM + Linux + Ollama
Resource group (RG) Container for related resources; delete the RG = delete everything in it 48 - Azure
Resource provider Azure service namespace that must be registered (Microsoft.App) 48 - Azure
REST Style of API built around resources and HTTP methods 09 - HTTP and APIs
Result backend Where job results / status are stored 42 - Redis, Caching and Task Queues
Retriever Component that finds relevant chunks for a query 31 - RAG (Retrieval-Augmented Generation)
Retriever / query engine Component that fetches documents for RAG 33 - Agent Frameworks
Reverse proxy Server forwarding client requests to backend apps 45 - Nginx, Reverse Proxy and HTTPS
root / sudo The administrator account / run one command as administrator 04 - Linux
Rubric Explicit criteria and scale for grading 35 - Evals and Observability
Runner The machine executing a job (GitHub-hosted or self-hosted) 44 - GitHub Actions (CI/CD)
Runtime Everything a program needs while it runs: the interpreter, installed packages, environment variables 01 - Core Concepts: Code, Packages, APIs and SDKs

S

Term Meaning Guide
safetensors Safe, fast file format for weights 25 - Hugging Face
Scaffolding Generating the initial files and folders of a project from a template 51 - Project Templates
Schema Rules describing valid structure (JSON Schema, Pydantic model) 08 - YAML, JSON, TOML and .env
Script A .py file you run 10 - Python Basics
SDK Software development kit: the official package that makes an API easy to use from one language 01 - Core Concepts: Code, Packages, APIs and SDKs
Secret Encrypted value (API key) available to workflows 44 - GitHub Actions (CI/CD)
Secrets API keys provided to the app securely, not in code 39 - AI User Interfaces (Streamlit, Gradio, Chainlit)
Semaphore A limit on how many tasks may run a section at once 14 - Async Python
Serialization Turning a model back into a dict / JSON 13 - Pydantic
Serialize / dump Python object -> text (json.dumps) 08 - YAML, JSON, TOML and .env
Series One column (or row) with an index 18 - Pandas
Server The program that receives requests and answers them 01 - Core Concepts: Code, Packages, APIs and SDKs
Server Program exposing tools / resources / prompts 34 - MCP (Model Context Protocol)
Server block Nginx config for one site / domain 45 - Nginx, Reverse Proxy and HTTPS
Server tools Run by the provider (web search, code execution) 29 - Tool Use (Function Calling)
Service Stable network name / IP that load-balances to pods 46 - Kubernetes
Service principal An identity for automation / CI pipelines 48 - Azure
Session Stored conversation state for an agent 33 - Agent Frameworks
Session state Per-user memory that survives reruns 39 - AI User Interfaces (Streamlit, Gradio, Chainlit)
SFT Supervised fine-tuning on (input, ideal output) examples 37 - Fine-tuning
Shape Size of each dimension, e.g. (3, 4) = 3 rows, 4 columns 17 - NumPy
Shell Program that interprets commands (PowerShell, CMD, Bash) 03 - Terminal and PowerShell
SKU / tier Size and price level of a resource (Basic, Standard, B1, S0) 48 - Azure
Source code The text files you write (app.py) that describe what a program should do 01 - Core Concepts: Code, Packages, APIs and SDKs
Space Hosted demo app (Gradio / Streamlit) 25 - Hugging Face
SSE Server-Sent Events: server streams chunks over one response (LLM streaming) 09 - HTTP and APIs
SSH Secure remote login to another machine 04 - Linux
SSH key Private / public key pair used instead of a password 49 - Azure VM + Linux + Ollama
SSH tunnel Forwarding a local port through SSH to a port on the VM 49 - Azure VM + Linux + Ollama
Standard library The modules that come with Python itself (json, pathlib, datetime) 01 - Core Concepts: Code, Packages, APIs and SDKs
State Data passed between steps (messages, intermediate results) 33 - Agent Frameworks
State File recording managed resources and their IDs 47 - Terraform
Status code 3-digit result: 2xx ok, 4xx client error, 5xx server error 09 - HTTP and APIs
Status code Result number: 200 OK, 404 not found, 422 invalid input 40 - FastAPI
Step One command (run) or one reusable action (uses) 44 - GitHub Actions (CI/CD)
Stop reason Why generation ended 27 - LLM APIs
Streaming Receive tokens as they are generated (SSE) 27 - LLM APIs
Streaming Showing tokens as they arrive 39 - AI User Interfaces (Streamlit, Gradio, Chainlit)
Strict mode No coercion: types must match exactly 13 - Pydantic
Strict tool use Guarantees arguments match the schema exactly 29 - Tool Use (Function Calling)
Structured output Response guaranteed to match a JSON schema 27 - LLM APIs
Sub-agent Agent started by another agent for a sub-task, with its own context 32 - AI Agents
Subscription Billing unit with its own limits and permissions 48 - Azure
Sync Make .venv match the lock file exactly 12 - uv
systemd / service Starts and supervises background programs (web server, Ollama) 04 - Linux
systemd service Keeps your app running and restarts it on failure / reboot 45 - Nginx, Reverse Proxy and HTTPS
System prompt Standing instructions for the whole conversation 28 - Prompt Engineering

T

Term Meaning Guide
Table / row / column Data as a grid: records and fields 20 - SQL
Target (y) The column to predict 23 - Scikit-learn
Task A coroutine scheduled to run concurrently (asyncio.create_task) 14 - Async Python
Task What a model does: text-classification, summarization, text-generation ... 25 - Hugging Face
Temperature Randomness of token choice 26 - LLM Fundamentals
Template version A Git tag on the template repository (for example v1.2.0) that copier update moves between 51 - Project Templates
Tenant Your organisation's directory (Microsoft Entra ID) 48 - Azure
Tensor N-dimensional array (like NumPy) that can live on CPU or GPU 24 - PyTorch
Terminal / shell A text window where you type commands to run programs 01 - Core Concepts: Code, Packages, APIs and SDKs
Test function def test_something(): with assert statements 15 - pytest
TLS / SSL certificate Proves the domain identity and enables encryption 45 - Nginx, Reverse Proxy and HTTPS
Token Chunk of text (word piece) the model reads and writes; ~4 characters of English 26 - LLM Fundamentals
Tokenizer Converts text to token IDs and back 25 - Hugging Face
Tokens per second (tok/s) Generation speed 36 - Local and Self-Hosted LLMs
Tool A command-line program (ruff, black) installed separately from projects 12 - uv
Tool A function the model may call, described by name + description + input schema 29 - Tool Use (Function Calling)
Tool A function the agent can call (29) 32 - AI Agents
Tool Callable function (model-controlled) 34 - MCP (Model Context Protocol)
tool_choice Whether / which tool the model must use 29 - Tool Use (Function Calling)
Tool definition The JSON description you send in tools=[...] 29 - Tool Use (Function Calling)
Tool loop / agentic loop Repeat: model -> tools -> model until no more tool calls 29 - Tool Use (Function Calling)
tool_result block Your answer, sent back with the matching tool_use_id 29 - Tool Use (Function Calling)
Tool runner SDK helper that runs the loop for you 29 - Tool Use (Function Calling)
tool_use block The model's request: tool name, input arguments and an id 29 - Tool Use (Function Calling)
Top-k Number of chunks retrieved 31 - RAG (Retrieval-Augmented Generation)
Trace / span Record of one request / of one step inside it 35 - Evals and Observability
Training / test set Data to learn from / data held back to check performance 23 - Scikit-learn
Transitive dependency A package your dependencies need (numpy for pandas) 12 - uv
Transport How messages travel: stdio (local process) or Streamable HTTP (remote server) 34 - MCP (Model Context Protocol)
TTL / expiry Seconds until Redis deletes a key automatically 42 - Redis, Caching and Task Queues

U

Term Meaning Guide
Upstream / backend The app behind the proxy 45 - Nginx, Reverse Proxy and HTTPS
Usage Input / output / cached token counts 27 - LLM APIs
User prompt The specific request / input for this turn 28 - Prompt Engineering
uv.lock Exact versions of every package; commit it to Git 12 - uv
uvloop / httptools Faster event loop and HTTP parser (included in uvicorn[standard]) 41 - Uvicorn (ASGI Server)
uv workspace Several Python packages in one repository sharing one uv.lock and one virtual environment 50 - Project Structure: Python Microservices + Frontend

V

Term Meaning Guide
Validation Checking and converting input data 13 - Pydantic
Validator Your own function that checks or transforms a field 13 - Pydantic
Variable / output Input parameter / value exported after apply 47 - Terraform
Vectorisation Applying an operation to a whole array at once 17 - NumPy
Vectorised operation Works on a whole column at once, no loop 18 - Pandas
venv Built-in Python module that creates environments 11 - Python Virtual Environment
.venv The environment folder in your project 11 - Python Virtual Environment
Version / semantic versioning A package's release number, MAJOR.MINOR.PATCH; a MAJOR change may break your code 01 - Core Concepts: Code, Packages, APIs and SDKs
Virtual environment A private folder of packages for one project, so projects do not conflict 01 - Core Concepts: Code, Packages, APIs and SDKs
VM size CPU / RAM / GPU combination you rent (e.g. Standard_D2s_v5) 49 - Azure VM + Linux + Ollama
Volume Persistent storage that survives container removal 43 - Docker
Volume / PVC Storage mounted into pods / a request for persistent storage 46 - Kubernetes
VRAM GPU memory; the main limit for GPU inference 36 - Local and Self-Hosted LLMs

W

Term Meaning Guide
Web API An API reached over HTTP, usually with JSON requests and responses 01 - Core Concepts: Code, Packages, APIs and SDKs
Webhook The server calls YOUR URL when something happens 09 - HTTP and APIs
Widget UI element (button, text input, file uploader) 39 - AI User Interfaces (Streamlit, Gradio, Chainlit)
Worker Separate process running the app; one per CPU core is a common start 41 - Uvicorn (ASGI Server)
Worker / consumer Process that runs jobs 42 - Redis, Caching and Task Queues
Workflow Fixed, code-defined sequence of LLM calls 32 - AI Agents
Workflow A YAML file describing automation 44 - GitHub Actions (CI/CD)
Working directory The folder the terminal is currently "in" 03 - Terminal and PowerShell
Workspace The folder (project) you have open 06 - VS Code
Workspace member One package inside a uv workspace (a service or a shared library) 50 - Project Structure: Python Microservices + Frontend
WSGI Older sync interface (Flask, classic Django) 41 - Uvicorn (ASGI Server)

Z

Term Meaning Guide
Zero-shot Instructions only, no examples 28 - Prompt Engineering