> ## Documentation Index
> Fetch the complete documentation index at: https://docs.pawa-ai.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Glossary

> A reference of key terms and concepts used throughout Pawa AI or AI industry in general

Here you’ll find clear explanations of important terms to help you understand AI ecosystem.

## SLM (Small Language Model)

SLMs (Small Language Models) are AI models that are trained on relatively smaller datasets and have fewer parameters compared to LLMs. They are lightweight, cost-efficient, and optimized for tasks that don’t require the full capacity of a large-scale model. SLMs are particularly useful for edge deployments, latency-sensitive applications, or environments with limited computational resources. While they may not have the broad generalization power of LLMs, SLMs excel in focused domains or task-specific scenarios.

## Embedding

Embeddings are vectorial representations of text, images, or other data types. They map complex data into a high-dimensional vector space where semantically similar inputs are located close together. For example, the words *king* and *queen* will have similar embeddings. Embeddings are crucial for tasks like semantic search, clustering, and classification. In Pawa AI, embeddings are stored in a vector database and power retrieval-augmented generation (RAG), recommendation systems, and knowledge discovery.

## Agents

Agents are AI-driven systems designed to perform autonomous tasks by combining reasoning, planning, and action. They use an LLM or SLM as a reasoning engine, often enhanced with memory, tools, and external APIs. Agents can handle multi-step workflows such as booking appointments, answering domain-specific questions, or orchestrating business processes. They can be reactive (responding to prompts) or proactive (taking initiative based on goals and triggers).

## Knowledge Base

A knowledge base is a structured repository of information — such as documents, FAQs, manuals, or articles — used to improve the accuracy of AI systems. When paired with embeddings and vector search, knowledge bases enable retrieval-augmented generation (RAG). This allows models to ground their responses in authoritative, up-to-date information instead of relying solely on pretraining data. Pawa AI supports uploading and managing knowledge bases that seamlessly integrate into workflows.

## Vectors

Vectors are numerical arrays representing data in a continuous high-dimensional space. In AI, they are used to encode semantic information about text, audio, or images. For example, a sentence like *“AI is powerful”* might be represented as a 1024-dimensional vector. Similar sentences will have vectors that are geometrically close. These properties make vectors essential for nearest-neighbor searches, recommendations, and similarity comparisons.

## Context Window

The context window refers to the maximum number of tokens (input + output) that a model can process at once. A larger context window allows the model to remember more information during a conversation or task. For example, with a 128k token context window, a model can handle entire books or large datasets in a single prompt. However, larger context windows require more computation and memory.

## Fine-tuning

Fine-tuning is the process of adapting a pre-trained model to a specific dataset, task, or domain. Instead of training from scratch, the model’s existing weights are adjusted slightly to learn task-specific patterns. Fine-tuning can improve accuracy, enforce domain-specific language, and reduce errors in specialized use cases. Variants include **instruction tuning** (aligning responses to follow directions better) and **parameter-efficient fine-tuning (PEFT)** methods such as LoRA, which reduce resource costs.

## RAG (Retrieval-Augmented Generation)

Retrieval-Augmented Generation (RAG) combines the power of LLMs with information retrieval systems. First, relevant documents or data are retrieved from a vector store using embeddings. Then, that information is injected into the model’s prompt to improve generation accuracy. RAG reduces hallucinations and ensures responses are factually grounded in external knowledge. It is widely used for chatbots, customer support, and enterprise knowledge management.

## Temperature

Temperature is a parameter that controls the randomness of model outputs. A lower temperature (e.g., 0.1) produces more deterministic and predictable results, useful for factual or structured tasks. A higher temperature (e.g., 0.9) increases creativity and variety, making it better for brainstorming or storytelling. Adjusting temperature lets developers balance precision and creativity depending on the application.

## Function Calling

Function calling allows a model to interact with external tools, APIs, or functions. Instead of just generating text, the model can decide to call a structured function, e.g., `get_weather("Dar-es-salaam")`. The API executes the function and returns the result, which the model then incorporates into its response. This makes AI systems more powerful, as they can combine natural language reasoning with deterministic external operations, such as retrieving real-time data, making calculations, or triggering business workflows.

### Observability

The ability to monitor, trace, and understand the internal state of a system through metrics, logs, and dashboards.

***