Bearning

09:00 – 10:30 | How Modern Language Models Work and Prompting Basics

• How a model processes a request: tokens, context and the objective function; why models hallucinate, and when a smaller, faster model is enough instead of the largest one — in terms of performance, latency and cost.
• Overview of today's leading models and AI assistants (ChatGPT, Claude, Gemini, Microsoft Copilot, Mistral, Llama and others) and how to choose the right one for the task.
• Reliable answers: how to request sources, ensure a consistent output style and set up the model once for repeated use (custom instructions, projects, custom assistants).
• Prompting basics: what distinguishes a request that produces usable output from one that leads to a generic or unusable answer.
• Working with your own documents — introduction to the afternoon block.

Interactivity:

• Polls and quizzes.
• Sharing participants' questions and expectations.

10:45 – 12:00 | AI Use Cases in Banking and Advanced Prompting

• Case studies: digital onboarding, document processing from scan to structured record, AML alert triage, fraud indicators, early warning systems and personalized advisory.
• The value of AI in practice: how to measure time savings and the level of automation.
• Advanced prompting in banking practice: the RACEF and CLEAR frameworks, when a plain instruction is enough and when to add examples (few-shot), how to request output in a precisely defined format, and how to build a reusable master prompt.

Interactivity:

• Mini-lab #1: Summarizing a long document, extracting key data and converting it into a table — using a document type from participants' own work.
• Mini-lab #2: Building your own master prompt for a recurring task.
• Polls and quizzes.

13:00 – 15:00 | RAG on Internal Data, Enterprise AI Assistants and Output Verification

• How AI works with internal documents through RAG: where documents are stored, how relevant content is retrieved, why the way documents are split matters, and how to request source citations.
• Key concepts explained: vector database, chunking and re-ranking.
• RAG vs. fine-tuning: when to use your own data and when to further train the model.
• Enterprise AI assistants in banking practice (Microsoft 365 Copilot, ChatGPT Enterprise, Claude, Gemini for Google Workspace, open-source and on-premise models): capabilities, limitations, data protection and choosing the right setup for the institution.
• Hallucinations: how to recognize them and how to verify facts before using the output further.

Interactivity:

• Mini-lab #3: Prompts that write prompts (meta-prompting) — using AI to create and refine instructions.
• Mini-lab #4: Managing hallucinations — identifying errors in model answers and ways to prevent them.

15:00 – 15:30 | Wrap-up, Q&A and Discussion

• Working through specific tasks from participants' practice — participants can bring their own task, which we will analyze together.