A glossary of common terms in Conversational AI: 

This is for those who are tech savvy and understand some AI Jargon. If not please read our easy to read glossary.

 

A

• Agentic AI – AI capable of autonomous goal-setting, planning, and tool use to accomplish complex tasks.

• API (Application Programming Interface) – A way for software to communicate; used in tool integration and AI-agent actions.

• ASR (Automatic Speech Recognition) – Converts spoken input into text for further NLP processing.

• Autonomous Agent – A system that independently takes actions toward goals with minimal human intervention.

 

---

 

B

• Bias – Unintended unfairness or skew in model outputs, often inherited from training data.

• Bot – Short for chatbot or voicebot; an automated agent that engages with users through conversation.

• BM25 – A ranking function used in keyword-based document retrieval (common in RAG systems).

 

---

 

C

• Chatbot – A conversational system that interacts with users through pre-defined or AI-generated text responses.

• Chunking – Splitting large documents into smaller parts for indexing and retrieval.

• Conversational AI – Technologies that enable natural, human-like dialogue between machines and users.

• Context Window – The span of tokens (words or symbols) an LLM can consider at once.

 

---

 

D

• Dialogue Management – The logic and structure that controls the flow of conversation in a conversational AI system.

• Domain-Specific – Refers to AI systems trained for particular industries or topics (e.g., medical, legal).

 

---

 

E

• Embedding – A dense numerical representation of text used to measure semantic similarity in RAG systems.

• Entity Recognition (NER) – Identifying key information (names, dates, locations) from text.

• Episodic Memory (in agents) – Memory of specific interactions, helping agents maintain context over time.

 

---

 

F

• Few-shot Learning – Using a small number of examples to teach an LLM how to perform a task.

• Fine-tuning – Training a pre-trained model on specific data to adapt it for a new domain or task.

• Feedback Loop – A system of learning and adaptation where AI behavior is improved through input from users or outcomes.

 

---

 

G

• Goal-Oriented Agent – AI designed to reach specific outcomes through planning and task execution.

• Grounding – Linking AI responses to verified external data to increase accuracy (critical in RAG).

 

---

 

H

• Hallucination – When an AI generates inaccurate or fabricated information not grounded in reality.

• Hybrid Search – Combines semantic vector search with keyword-based techniques for better retrieval.

 

---

 

I

• Intent Recognition – Determining the user’s goal or desired action from their input.

• Inference – The act of generating an output (e.g., a response or prediction) using a trained model.

 

---

 

J

• JSON (JavaScript Object Notation) – A common format for structuring data in APIs, often used by agents to interact with tools.

 

---

 

K

• Knowledge Base – A structured or semi-structured repository of information that AI systems use for RAG.

• Knowledge Grounding – The process of ensuring AI outputs are informed by trusted sources.

 

---

 

L

• LLM (Large Language Model) – A type of AI model trained on massive datasets to understand and generate language.

• LangChain – A framework for developing applications powered by LLMs, useful for building RAG and agent workflows.

 

---

 

M

• Memory (Agent Memory) – Mechanisms for storing and recalling past interactions or knowledge.

• Multimodal AI – AI capable of understanding and generating across text, speech, image, or video modalities.

 

---

 

N

• Natural Language Processing (NLP) – A field of AI focused on understanding, interpreting, and generating human language.

• NLG (Natural Language Generation) – The process of producing human-like language as output.

• NLU (Natural Language Understanding) – The process of interpreting user input in natural language.

 

---

 

O

• Open-Domain Chatbot – A chatbot capable of handling unrestricted conversations across many topics.

• Ontology – A structured framework of knowledge used to support reasoning or organization in AI systems.

 

---

 

P

• Prompt Engineering – Crafting input prompts to guide LLM behavior effectively.

• Planning (in Agentic AI) – Creating a sequence of actions to achieve a defined goal.

• Post-Retrieval Reranking – Improving result quality by re-evaluating retrieved documents based on relevance.

 

---

 

Q

• Query Expansion – Adding related terms to improve retrieval results in search and RAG pipelines.

 

---

 

R

• RAG (Retrieval-Augmented Generation) – A hybrid approach combining external document retrieval with LLM-based generation.

• Retriever – The component in RAG that fetches relevant documents based on a user query.

• Reinforcement Learning (RLHF) – Training a model using feedback on its outputs to improve performance.

• Reflection (in agents) – The ability of agents to analyze their own behavior and improve future performance.

 

---

 

S

• Semantic Search – Search based on meaning or context rather than exact keywords, powered by embeddings.

• Slot Filling – Collecting necessary pieces of information in dialogue to complete a task (e.g., name, date, location).

• Speech Synthesis (TTS) – Converting text into spoken audio.

 

---

 

T

• Task Decomposition – Breaking down a large goal into smaller subtasks, critical for agent execution.

• Tool Use (in Agentic AI) – The ability of an agent to call APIs, run code, or use databases to accomplish tasks.

• Transformer – The neural network architecture underlying most modern LLMs.

 

---

 

U

• Utterance – A single input from a user in a conversation (spoken or written).

• User Simulation – Artificially generating user input for training or testing dialogue systems.

 

---

 

V

• Vector Database – A specialized database for storing and querying embeddings, essential for RAG systems.

• Voicebot – A conversational AI system that operates via speech rather than text.

 

---

 

W

• Workflow Automation (Agentic AI) – Using autonomous agents to complete multi-step tasks or business processes.

 

---

 

X

• Explainability (XAI) – Techniques to make AI decisions understandable to humans, crucial in regulated industries.

 

---

 

Y

• Yield Optimization (Conversational) – Improving the conversion or effectiveness of AI-led conversations (e.g., in sales).

 

---

 

Z

• Zero-shot Learning – Performing a task without specific prior examples, relying on general knowledge from pre-training.