98 - Glossary¶
Previous: 97 - Capstone Project: Document Chatbot | Index: All guides | Next: 99 - Quick Reference
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 |
Previous: 97 - Capstone Project: Document Chatbot | Index: All guides | Next: 99 - Quick Reference