AI, Agentic & BIM Glossary — plain-English definitions

AI (Artificial Intelligence)

Computer systems that perform tasks normally requiring human intelligence — recognising images, understanding language, making decisions.

Deep Learning

Machine learning using neural networks with many layers — the basis of modern AI.

Machine Learning (ML)

Algorithms that learn patterns from data instead of being explicitly programmed.

Neural Network

A computational system loosely inspired by the brain — layers of nodes that transform input into output.

Training Data

The labelled examples used to teach a machine learning model what to do.

Chain of Thought (CoT)

A prompting technique where the LLM is asked to reason step-by-step before producing an answer.

Chunking

Splitting documents into smaller passages before indexing them for RAG retrieval.

Context Window

The maximum number of tokens an LLM can process in a single prompt + response.

Embedding

A numerical representation of text, images, or other data — used to measure semantic similarity.

Fine-tuning

Continuing to train a pre-trained model on a specific dataset to adapt it to a task.

Hallucination

When an LLM generates content that is fluent and confident but factually wrong.

Inference

Running a trained model to produce predictions — the production-time work, as opposed to training.

LLM (Large Language Model)

A neural network trained on huge amounts of text — capable of understanding and generating human language.

LoRA (Low-Rank Adaptation)

A fine-tuning technique that trains a small number of new parameters instead of updating the full model.

Multimodal

A model that processes more than one input modality — typically text + images, sometimes audio or video.

Prompt Caching

A vendor feature that caches the prefix of a prompt so repeated calls reuse it at a discount.

Prompt Engineering

The discipline of crafting LLM inputs that reliably produce useful outputs.

Quantization

Compressing a model by representing its weights with fewer bits — typically 8-bit or 4-bit instead of 16-bit.

RAG (Retrieval-Augmented Generation)

A pattern where an LLM retrieves relevant documents before answering — making it factual and up-to-date.

Reranking

A second-stage scoring of retrieved chunks using a stronger model to improve RAG precision.

RLHF (Reinforcement Learning from Human Feedback)

The post-training technique that turns a raw LLM into a helpful, harmless, instructable assistant.

Structured Output

Forcing an LLM to return valid JSON (or a defined schema) instead of free-form text.

Token

The basic unit of text an LLM processes — typically a word fragment, not a full word.

Transformer

A neural network architecture introduced in 2017 — the foundation of every modern LLM.

Vector Database

A specialised database that stores embeddings and efficiently finds nearest neighbours.

Agent Memory

Persistent state an agent carries across turns or sessions — short-term context, long-term facts, episodic history.

Agentic AI

AI systems that take goal-directed action — planning, using tools, and adapting — rather than just answering questions.

Claude Code

Anthropic's agentic coding tool — Claude with file-edit, command-run, and project-context tools, running in your terminal or IDE.

Evaluation (Evals)

Systematic measurement of LLM and agent quality — the production discipline that distinguishes engineering from prompting.

Function Calling

The LLM feature that lets the model decide to invoke a named function with structured arguments.

Guardrails

Programmatic checks around LLM inputs and outputs — what the model can be asked, what it is allowed to return.

Multi-Agent System

A system where several specialised agents collaborate — each with a defined role, sharing state via messages or memory.

ReAct (Reason + Act) Pattern

The canonical agent loop where the LLM alternates between reasoning steps and tool-use steps.

Tool Use

The pattern where an LLM invokes external tools (APIs, functions, code) to take action beyond text generation.

AI Agent

An LLM-powered system that plans, uses tools, and completes multi-step tasks autonomously.

Claude API

The API for accessing Anthropic's Claude models programmatically.

Hugging Face

A platform + library hub for open-source machine learning models and datasets.

LangChain

An open-source framework for building LLM-powered applications in Python and JavaScript.

LangGraph

A LangChain extension for building stateful, multi-agent workflows as graphs.

MCP (Model Context Protocol)

An open standard for connecting LLMs to external tools and data sources.

OpenAI API

The API for accessing OpenAI's GPT models programmatically.

Classification

An ML task that predicts a discrete category — spam vs not-spam, churn vs no-churn.

Cross-Validation

A technique to estimate how well an ML model will generalise to unseen data.

ETL (Extract, Transform, Load)

The process of moving data from source systems into a data warehouse, cleaning it along the way.

Feature Engineering

The process of creating input variables (features) for ML models from raw data.

Regression

An ML task that predicts a continuous number — sales, salary, temperature.

Data Analyst

A professional who turns business data into decisions via SQL, dashboards, and statistical analysis.

Data Scientist

A professional who builds predictive models that drive business decisions.

ML Engineer

A software engineer specialised in deploying and operating ML systems in production.

MLOps Engineer

A specialised ML Engineer focused on infrastructure, CI/CD, monitoring, and reliability for ML systems.

AutoCAD

Autodesk's 2D drafting software — still essential alongside Revit for legacy work and detailed drawings.

BIM 360 / Autodesk Construction Cloud

Autodesk's cloud platform for coordinating BIM projects across distributed teams.

Clash Detection

The process of finding physical conflicts between disciplines in a federated BIM model before they hit site.

Dynamo for Revit

A visual programming language that automates repetitive tasks in Revit — the senior BIM differentiator.

IFC (Industry Foundation Classes)

The open vendor-neutral file format for exchanging BIM models between different software.

ISO 19650

The international standard for organising and managing information through the lifecycle of built assets using BIM.

LOD (Level of Development)

A measure of how reliable and detailed each element of a BIM model is at a given project stage.

Navisworks

Autodesk's coordination + clash detection tool — where federated BIM models are checked for conflicts.

Revit

Autodesk's flagship BIM authoring software — the dominant tool for modelling buildings in India + the Middle East.