Napojiť sa či navrhnúť? Dve cesty k umelej inteligencii – a prečo v MAYAUY volíme tú druhú

4. Hybrid, but on our terms

At MAYAUY, we recognize that not everything needs to be reinvented from scratch. In some cases, we can effectively leverage a small part of an existing model — for example, Microsoft Azure models or other LLM tools — as a starting layer, on top of which we build our own solution. We mainly use them as a generative or interface engine, while:

• all business data logic,

• prompt architecture and decision layers,

• and the user experience

  …are designed, adapted, and optimized for the specific client.

In other words, we don’t use the existing model as the core, but as technical assistance. The control over what the AI sees, processes, and how it decides remains on our side.

This is how we create solutions that have the power of global technologies, but the personalization of a tailored suit. And that’s the key difference between just “plugging in” and building an AI solution responsibly and meaningfully.

5. Differences between a custom AI model and connecting to an existing AI model

5.1. Custom AI model (“custom AI”)

Creating a custom AI model means that the algorithm is trained specifically on your data for a particular purpose or problem. In practice, it means developing a “custom AI model.”

Advantages:

• Domain-specific accuracy: The model “understands” the specifics of your business, language, customers, documents, etc.

• Full control over architecture and output: You set how the model makes decisions.

• Ownership and GDPR compliance: Both data and model remain in your environment.

Disadvantages:

• Higher costs for development, testing, and maintenance

• Requires quality data and a data science team

• Longer deployment time

Using a custom model is suitable for legal departments, insurance, healthcare, or manufacturing companies — anywhere the data is specific and sensitive.

5.2. Connecting to an existing AI model (“plug & play”)

This approach uses existing large language or visual models (such as GPT, Claude, Gemini, DALL·E) via API or integrated services. The model is not retrained from scratch, but customized via “prompt engineering” or fine-tuning.

Advantages:

• Faster deployment

• Low initial costs and minimal infrastructure needs

• Access to state-of-the-art models without the need for development

Disadvantages:

• Limited ability to explain or adjust the model’s decision-making

• Risk of data leakage (if not technically secured properly)

• Dependence on the service provider (licenses, availability, outages)

This approach is ideal for smaller teams, startups, or companies that need to quickly validate an idea and don’t have enough specific data to train their own model.

When to use which?

If you’re looking for a quick MVP or prototype, choose an existing model.

If you need a sensitive, highly customized solution, it pays to invest in your own model.

At MAYAUY, we often design hybrid approaches — where the base is an existing model, and on top of it, we build a thin “layer of your identity” through API interfaces, custom data, or prompt scenarios.

6. Legal Differences: Custom AI vs. API-Based Systems

From a legal perspective, there are several important differences between a custom AI system and an AI system using an API, especially regarding liability, ownership, data protection, and licensing conditions.

6.1. Custom AI System

This system is designed and developed specifically for a particular client:

• Ownership and IP: Intellectual property rights can be transferred to the client if the contract allows. In the case of in-house development, the rights usually remain with the company.

• Liability: Liability for the system’s outputs can usually be contractually regulated – the developer or client can agree on guarantees and liability limitations.

• Data Protection: The processing of personal data is directly under the control of the system owner, which can make fulfilling GDPR requirements easier.

6.2. API-Based AI System (e.g., via a third-party provider)

Here, the client only uses a ready-made solution available as a service:

• Ownership and IP: The developer does not have access to the source code and does not own the model. The outputs may be licensed, but not necessarily owned by the client.

• Liability: Liability is limited by the API provider’s contract (e.g., OpenAI, Microsoft), and companies often exclude legal responsibility for outputs in their terms of service.

• Data Protection: Data transmitted via the API may be processed according to the provider’s policies, which requires thorough analysis of contractual terms and data protection policies.

In practice, this means that a custom system offers greater control and customization, while an API solution provides rapid deployment with limited legal impact on the model itself.

Conclusion

Connecting to an existing AI may be tempting, but it’s not enough in key areas. If AI is to be a truly strategic partner for your business, it must be yours—trustworthy, understandable, and adaptable.

That’s why at MAYAUY, we focus on developing custom and hybrid AI models. Because we believe technology should adapt to the company—not the other way around.