# 4. Architecture

#### 4.1 High Level System Architecture

MAYA does not follow a single model centric structure. Instead, it adopts a template driven multimodal architecture.

The overall flow of the system is structured as follows:

> \[Contributor Layer]
>
>         ↓ (Template / Whitebox)
>
> \[Template Registry]
>
>         ↓
>
> \[Context Assembly Layer]
>
>         ↓
>
> \[Multimodal Generation Engine]
>
>         ↓
>
> \[Output Delivery Layer]

The design intent of this architecture is clear.

* Generation standards are defined by humans, not by the model
* AI is focused on execution rather than decision making

This structure allows MAYA to maintain consistency in output quality while keeping AI usage efficient and predictable.

***

#### 4.2 Template and Whitebox Layer (Human Intelligence Layer)

This layer represents the core differentiation of MAYA. Trends, meme structures, tone and format, and cultural context are stored as reusable templates, referred to as whiteboxes.

By structuring these elements at the template level, MAYA achieves the following advantages:

* Output quality does not fluctuate randomly
* Cultural understanding does not rely solely on model training
* Trends become long term assets rather than one time data

As a result, human insight is preserved and reused, instead of being lost after a single campaign or trend cycle.

***

#### 4.3 Context Assembly Layer (Cost Control Layer)

The context assembly layer combines user inputs, company data, and template information, then delivers only the minimum required context to the AI.

This design enables:

* Reduced unnecessary token consumption
* Prevention of redundant model calls
* Improved predictability of generation costs

By tightly controlling context size and structure, MAYA ensures stable performance without sacrificing output quality.

***

#### 4.4 Multimodal Generation Engine

MAYA is not bound to a single model. It selectively generates images, videos, text, or audio depending on the use case and objective.

The key principle is that decisions about what to generate are already made at the upper layers of the system. The generation engine focuses purely on execution.

This results in:

* Reduced decision burden on the model
* Consistent output quality across content types
* A structure that allows easy expansion into new modalities

***

#### 4.5 Korean First Localization Pipeline

Korean is treated as a primary output language, not as a translated result. MAYA maintains a dedicated pipeline that accounts for sentence length, tone, readability, and visual layout.

This enables:

* Content that can be deployed immediately in the Korean market
* No additional editing required for CS, announcements, or marketing content
* Significant reduction in localization costs for global companies

***

#### 4.6 CS Optimized Architecture (Enterprise Layer)

In customer support use cases, MAYA does not route all requests through AI.

The CS architecture is structured as follows:

> \[FAQ and rule based layer]\
> ↓\
> \[Template based response layer]\
> ↓\
> \[Selective AI invocation layer]

This layered approach delivers the following benefits:

* Prevention of sudden spikes in AI usage
* Predictable monthly operating costs
* Consistent response quality that matches Korean tone and expectations

Through this architecture, MAYA provides enterprise grade customer support optimization while maintaining both cost efficiency and localization quality.

<br>


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://maya-ai-2.gitbook.io/maya.ai-docs/official/en/4.-architecture.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.