Conference

Don't ask whether your AI agent is intelligent. Ask what it knows

At AI Summit Poland 2026, we presented what determines the quality of AI responses in practice: data, its organization, freshness, semantics, and accessibility. Your AI knows exactly as much as your data can tell it — and it does not know what it does not know.

AI Summit Poland 2026 | Data organization as a prerequisite for effective LLM systems

Back to events

Don't ask whether your AI agent is intelligent. Ask what it knows

What was the presentation about?

What does your AI really know?

A large language model does not query your organization's systems on its own. To answer a question about the company, it must receive the right context or gain access to data through tools. This means that the quality of an AI response begins before the model itself — it begins with data. We showed what happens between a user's question and the model's response, and why having complete data does not yet mean that AI can use it effectively.

What did we show?

How an LLM works and where organizational data enters: We explained where company data enters the model's process, how knowledge stored in the model differs from knowledge supplied in context, and why a model can give a convincing answer even without a complete picture.
Where the data AI needs actually resides: Organizational data is distributed across source systems, data warehouses, data lakes, documents, and knowledge repositories. Some of the most important context lies elsewhere — in decisions, exceptions, and human experience. We showed why a model needs not only data itself, but also its meaning, freshness, version, owner, and appropriate permissions.
Why RAG does not solve everything: RAG is now one of the most common ways to provide knowledge to models, but retrieving a similar passage of text is not always enough. We showed when RAG works well, when it begins to fail, and why many RAG problems are actually data-organization problems.

What can replace or complement RAG?

We also presented other approaches to providing models with knowledge:

Knowledge graphs and GraphRAG

Semantic layers and text-to-SQL

Agents using APIs and MCP

Data products and feature stores

Context from events

Skills and knowledge repositories for agents

The goal is not to choose one technology. It is to match the architecture to the type of data and the question AI is expected to answer.

Six questions to ask before implementing AI

Is there one definition and a designated source for each key business concept?

Do you know how current the data behind the model's answer is?

Do source-system permissions carry through to indexes and model responses?

Can you identify the source and version of the data used to produce the answer?

Can the model recognize that it does not have enough information?

And finally: does the organization know what its AI does not know?

Data is the foundation of AI

AI architecture is secondary to data organization. RAG, knowledge graphs, and agents can use data only to the extent that it is accessible, described, current, and properly secured. Preparing data for AI is therefore not the final stage of an implementation. It is the foundation of the entire solution. During the presentation, we examined the problem from a data-architecture perspective — from source systems and semantics, through RAG and agents, to the architecture of the AI ecosystem.

Speaker

Piotr Boguszewski

Piotr Boguszewski has worked with data and analytics for nearly 20 years, from data warehouses to CRM systems. He focuses on connecting business and technology perspectives and translating complex data challenges into practical solutions. In recent years, he has concentrated primarily on artificial intelligence, particularly Generative AI, and on putting its potential to practical use in organizations. His experience spans both consulting and team management, from small specialist teams to larger regional structures.

Speaker

Jakub Fila

An experienced systems and solutions architect specializing in application design, efficient programming, data warehousing, big data, numerical computing, and machine learning algorithms. He studied Applied Mechanics and Physics / Applied Mathematics and completed an MBA in management accounting and finance at Warsaw University of Technology. His career includes IBM, Teradata, Accenture, Deloitte Digital, and BCG Platinion, combining deep technical expertise with strong business understanding.

Want to getthe presentation or access?

Phone

+48 22 208 28 26

Email address

biuro@goldenore.com

Office

Al. Jerozolimskie 160, 02-326 Warsaw

Company details

Goldenore Sp. z o.o.

KRS: 0000924344 | NIP: 6342815553 | REGON: 243185818

Watch the recording

Watch the full presentation, “What Does Your AI Know? Data Organization as a Prerequisite for Effective LLM Systems,” from AI Summit Poland 2026 and see what must happen in the data layer before AI can truly use your organization's knowledge. Let's talk about the data behind AI in your company.

AI Summit Poland 2026

Photo gallery

Explore photos from the conference, our presentation, and the roundtable discussions.