AI Factory terminology
======================

AI Factory terminology defines key concepts and technologies used across
**EDB Postgres® AI** (EDB PG AI), **Hybrid Manager AI Factory**, and
related components.

This page complements these conceptual explanations:

- :ref:`Gen AI Builder Concepts (Assistants, Knowledge, Tools, Threads) <Gen AI Builder Concepts (Assistants, Knowledge, Tools, Threads)>`  

- :ref:`Fundamental AI Concepts <Fundamental AI Concepts>`  

Understanding these terms helps you build trusted, governed, and
scalable AI solutions with EDB PG AI.

``---``

Core AI concepts
----------------

Machine learning (ML)
^^^^^^^^^^^^^^^^^^^^^

Machine learning uses algorithms that improve through data exposure to
perform tasks without explicit programming. It powers predictive
analytics, automation, and decision making in AI Factory through
:ref:`Model Serving Reference Manual <Model Serving Reference Manual>`  , Pipelines, and Assistants.

`Learn more <https://en.wikipedia.org/wiki/Machine_learning>`_ 

- NOTE THIS IS NOT AN INTENDED USE CASE SUPPORTED BY HYBRID MANAGER
  CURRENTLY \*

Deep learning (DL)
^^^^^^^^^^^^^^^^^^

Deep learning leverages multi-layer neural networks to model complex
data patterns. It supports advanced applications such as language
understanding and image recognition within AI Factory, enabled through
:ref:`Model Serving Reference Manual <Model Serving Reference Manual>`  with GPU acceleration.

`Learn more <https://en.wikipedia.org/wiki/Deep_learning>`_ 

Natural language processing (NLP)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

Natural language processing (NLP) enables computers to understand and
generate human language. It underpins semantic search and conversational
AI in AI Factory via :ref:`Knowledge Bases Reference Manual <Knowledge Bases Reference Manual>`  , :ref:`Retrievers Reference Manual <Retrievers Reference Manual>`  , and :ref:`Assistants Reference Manual <Assistants Reference Manual>`  .

`Learn more <https://en.wikipedia.org/wiki/Natural_language_processing>`_ 

Large language models (LLMs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^

LLMs are large deep learning models trained on massive text corpora. In
AI Factory, they drive :ref:`Assistants Reference Manual <Assistants Reference Manual>`  , Retrieval-Augmented Generation
(RAG), and various model pipelines, deployed using :ref:`Model Serving Reference Manual <Model Serving Reference Manual>`  .

`Learn more <https://huggingface.co/blog/llm>`_ 

Embeddings
^^^^^^^^^^

Embeddings are vector representations of data that capture semantic
meaning. AI Factory Pipelines create embeddings used in :ref:`Knowledge Bases Reference Manual <Knowledge Bases Reference Manual>` 
and served through the :ref:`Vector Engine (PGvector) <Vector Engine (PGvector)>`  to enable semantic search and RAG.

`Learn more <https://sebastianraschka.com/blog/2023/llm-embeddings.html>`_ 

Vector databases
^^^^^^^^^^^^^^^^

Vector databases store embeddings and enable fast similarity search. AI
Factory provides this through the :ref:`Vector Engine (PGvector) <Vector Engine (PGvector)>`  , built on the
open-source `pgvector <https://github.com/pgvector/pgvector>`_  extension, integrated directly with Postgres.

Retrieval-augmented generation (RAG)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

RAG combines vector search with LLM generation to ground model responses
in relevant documents. In AI Factory, it is implemented through
:ref:`Knowledge Bases Reference Manual <Knowledge Bases Reference Manual>`  , :ref:`Retrievers Reference Manual <Retrievers Reference Manual>`  , and :ref:`Model Serving Reference Manual <Model Serving Reference Manual>`  .

`Intro to RAG <https://huggingface.co/blog/rag>`_ 

``---``

AI for databases
----------------

Intelligent database management
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

Intelligent database management applies AI to optimize Postgres
performance and operations. AI Factory extends this with intelligent
retrieval and search using :ref:`Vector Engine (PGvector) <Vector Engine (PGvector)>`  and Pipelines.

Vector search in Postgres
^^^^^^^^^^^^^^^^^^^^^^^^^

Vector search allows you to query embeddings directly within Postgres.
AI Factory uses `pgvector <https://github.com/pgvector/pgvector>`_  to power this capability through the
:ref:`Vector Engine (PGvector) <Vector Engine (PGvector)>`  , supporting Knowledge Bases and RAG.

AIDB
^^^^

AIDB (AI-in-Database) brings vector search, embedding pipelines, and
future capabilities to HM-managed Postgres clusters. It is the
foundation for AI Factory :ref:`Pipelines and data preparation <Pipelines and data preparation>`  and Knowledge Bases. ========
Intelligent database management applies AI to optimize Postgres
performance and operations. AI Factory extends this with intelligent
retrieval and search using `Vector Engine <https://www.enterprisedb.com/docs/edb-postgres-ai/latest/ai-factory/vector-engine/>`_  and Pipelines.

In-database machine learning (In-DB ML)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

In-DB ML enables running vector search and ML pipelines inside Postgres,
reducing data movement and latency. AI Factory implements this through
`Vector Engine <https://www.enterprisedb.com/docs/edb-postgres-ai/latest/ai-factory/vector-engine/>`_  and `Pipelines <https://www.enterprisedb.com/docs/edb-postgres-ai/latest/ai-factory/pipeline/>`_  .

.. _vector-search-in-postgres-1:

Vector search in Postgres
^^^^^^^^^^^^^^^^^^^^^^^^^

Vector search allows you to query embeddings directly within Postgres.
AI Factory uses `pgvector <https://github.com/pgvector/pgvector>`_  to power this capability through the
:ref:`Vector Engine (PGvector) <Vector Engine (PGvector)>`  , supporting Knowledge Bases and RAG.

.. _aidb-1:

AIDB
^^^^

AIDB brings vector search, embedding pipelines, and other capabilities
to HM-managed Postgres clusters. It is the foundation for AI Factory
`Pipelines <https://www.enterprisedb.com/docs/edb-postgres-ai/latest/ai-factory/pipeline/>`_  and Knowledge Bases.

Natural language interfaces to databases
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

Natural language interfaces enable users to query Postgres using natural
language rather than SQL. AI Factory Assistants leverage this pattern
through :ref:`Gen AI Builder <Gen AI Builder>`  , combining LLMs with structured and unstructured
retrieval.

``---``

AI infrastructure
-----------------

AI-accelerated hardware
^^^^^^^^^^^^^^^^^^^^^^^

AI Factory uses GPU-accelerated Kubernetes clusters to serve deep
learning models and high-throughput inference. Model workloads in
`Model Serving <https://www.enterprisedb.com/docs/edb-postgres-ai/latest/ai-factory/model/serving/>`_  run on GPU-enabled nodes.

`Learn more <https://developer.nvidia.com/cuda-gpus>`_ 

KServe
^^^^^^

KServe is the open-source Kubernetes-native framework AI Factory uses to
deploy and manage models. It provides InferenceServices, autoscaling,
and observability for AI Factory :ref:`Model Serving Reference Manual <Model Serving Reference Manual>`  .

`KServe documentation <https://kserve.github.io/website/>`_ 

Model Serving
^^^^^^^^^^^^^

Model Serving deploys AI models as production-grade inference services,
using KServe under the hood. It supports LLMs, embedding models, vision
models, and custom AI workloads.

See: :ref:`Model Serving Reference Manual <Model Serving Reference Manual>`  , :ref:`Model Serving Explained <Model Serving Explained>` 

``---``

Gen AI Builder
--------------

.. _gen-ai-builder-1:

Gen AI Builder
^^^^^^^^^^^^^^

Gen AI Builder is the application layer of AI Factory for building
AI-powered apps and agents. It integrates:

- :ref:`Assistants Reference Manual <Assistants Reference Manual>`  for conversational AI 

- :ref:`Knowledge Bases Reference Manual <Knowledge Bases Reference Manual>`  for RAG and semantic search 

- :ref:`Retrievers Reference Manual <Retrievers Reference Manual>`  for advanced search patterns 

- :ref:`Tools Reference Manual <Tools Reference Manual>`  to perform external actions 

- :ref:`Rulesets Manual <Rulesets Manual>`  to control Assistant behavior 

- :ref:`Threads Reference Manual <Threads Reference Manual>`  to persist conversation history 

- :ref:`Structures Reference Manual <Structures Reference Manual>`  to define AI workflows and pipelines 

It is deployed on :ref:`Hybrid Manager AI Factory <Hybrid Manager AI Factory>`  , enabling full-stack Sovereign AI.

``---``

Hybrid Manager AI Factory
-------------------------

Hybrid Manager
^^^^^^^^^^^^^^

Hybrid Manager provides the Kubernetes-based control plane for AI
Factory workloads, managing GPU resources, Pipelines, Model Serving, and
Knowledge Bases. It provides end-to-end governance, security, and
observability.

See: :ref:`Hybrid Manager AI Factory <Hybrid Manager AI Factory>` 

Pipelines
^^^^^^^^^

Pipelines automate the creation of embeddings and Knowledge Bases from
data sources in your control, ensuring transparency and auditability.

See: :ref:`AI Accelerator Pipelines overview <AI Accelerator Pipelines overview>` 

Knowledge Bases
^^^^^^^^^^^^^^^

Knowledge Bases index content for semantic search and RAG. They support
multi-source, multi-modal search and power AI Factory Assistants.

See: :ref:`Knowledge Bases Reference Manual <Knowledge Bases Reference Manual>` 

.. _model-serving-1:

Model Serving
^^^^^^^^^^^^^

Model Serving deploys models using Kubernetes-native KServe and
integrates with the :ref:`Model Library Reference Manual <Model Library Reference Manual>`  . It powers Assistants, Knowledge
Bases, and custom AI applications.

See: :ref:`Model Serving Reference Manual <Model Serving Reference Manual>` 

``---``

Additional concepts
-------------------

Image and Model Library
^^^^^^^^^^^^^^^^^^^^^^^

The Image and Model Library in Hybrid Manager manages container images
for both Postgres and AI model deployments. The :ref:`Model Library Reference Manual <Model Library Reference Manual>`  provides
an AI-focused view, supporting Model Serving and governed image
workflows.

See: :ref:`Model Library Reference Manual <Model Library Reference Manual>` 

NVIDIA NIM
^^^^^^^^^^

NVIDIA NIM Microservices are optimized AI model images from NVIDIA for
high-performance inference. AI Factory uses NIM containers for LLMs,
embeddings, and vision models, served through Model Serving.

See: :ref:`How-To Deploy an NVIDIA NIM Microservices available on build.nvidia.comcontainer with KServe <How-To Deploy an NVIDIA NIM Microservices available on build.nvidia.comcontainer with KServe>` 

``---``

Summary
-------

Understanding these terms will help you build with:

- :ref:`Gen AI Builder Concepts (Assistants, Knowledge, Tools, Threads) <Gen AI Builder Concepts (Assistants, Knowledge, Tools, Threads)>`  

- :ref:`AI Factory Explained <AI Factory Explained>`  

- :ref:`Hybrid Manager AI Factory <Hybrid Manager AI Factory>`  

Together, these capabilities enable you to deliver **trusted, governed,
Sovereign AI** — with your data, in your databases, under your control.

``---``