Practical AI Tools: Machine Learning, Deep Learning, and RAG for Analysts and Non-Programmers

1. How artificial intelligence works in practice

Modern AI is presented through the mechanisms that matter in practice: how models generate answers, why some responses are useful, and what data contributes to that process. The section also organizes key ML and Deep Learning concepts, model types, and core training notions relevant to analysts.

• what modern AI „consists of”,
• how the model generates a response,
• where good responses come from,
• basics of Machine Learning and Deep Learning:
  • supervised learning, unsupervised learning, reinforcement learning (business examples: scoring, segmentation, recommendations),
  • neural networks, deep networks, what distinguishes „classical” ML from Deep Learning,
• how Machine Learning works „under the hood” for an analyst:
  • data as the foundation of models: features, labels, the process of „learning from examples”,
  • concepts: training, validation, test, overfitting, underfitting,
• types of models: predictive, generative, LLMs:
  
  • predictive models (regression, classification) in business tasks,
  • generative models (text, images, code) vs „classical” predictive models,
  • reasoning models.

2. AI in data analysis on a daily basis

Daily analytical work often starts with repetitive tasks in Excel, data preparation, and quick interpretation of trends. Here the emphasis is on typical analyst scenarios, handling structured and unstructured data, and using tools such as Copilot, ChatGPT, Gemini, Excel, and Power BI as an AI support layer.

• supporting work in Excel,
• preparation and description of data,
• trend analysis and the first list of hypotheses,
• typical scenarios for an analyst,
• structured and unstructured data,
• analytical tools with AI without programming – Copilot, ChatGPT, Gemini as an „AI layer” over Excel, SQL, BI,
• tools of the type „data analysis with AI” in Excel/Power BI.

3. Can AI be trusted? Errors, biases and data quality in an analyst's work

Trust in AI depends on understanding where errors come from and what the data can actually support. The section addresses hallucinations, bias in models and datasets, the role of data quality in sound analysis, and practical boundaries of responsibility, ending with simple rules for safer daily use.

• hallucinations and errors – how to understand them,
• biases (bias) in models and data,
• data quality as a condition for meaningful analysis,
• frameworks of trust: what AI can (not) be responsible for,
• simple rules of „safe use” for the analyst.

4. Creating your own tools with the help of AI (extended with AutoML / no‑code)

Building analyst tools becomes broader when AI is combined with AutoML and no-code platforms. Participants work through translating a problem into templates, calculators, automations, macros, and queries, then extend that approach to predictive models, low-code environments, and AI-assisted data analysis tools.

• how to describe a problem so that AI can design a tool,
• file templates, calculators, checklists,
• simple automations, macros, queries,
• good practices in building „your own tools”,
• AutoML and predictive models without programming,
• low-code / no-code platforms for analysts,
• AI-powered data analysis tools.

5. RAG systems and working with company data (NotebookLM, MS Notebooks, AnythingLM)

Working with company knowledge requires more than a chat interface, especially when data is scattered across documents and datasets. This part explains how RAG works, how to prepare structured and unstructured sources, how to design effective queries, and how no-code tools support practical use.

• what RAG (Retrieval-Augmented Generation) is,
• data for RAG: structured, unstructured, quality problems,
• examples of using RAG without programming,
• designing queries for RAG systems,
• preparing data for working with RAG.

6. Fundamentals of Machine Learning in an analyst's practice

From a business question to model output, the focus here is on recognizing when prediction, classification, or recommendation is truly an ML task. It also explains a simple project lifecycle, feature preparation, data splits, and quality metrics, then contrasts classic ML with LLMs and RAG in analytical work.

• from the business problem to the model – how to recognize that this is an „task for ML”: prediction, classification, recommendation,
• how to translate a business question into the model's input/output:
  
  • life cycle of a simple ML project,
  • understanding data, feature preparation, split into sets, quality assessment,
  • metrics: accuracy, precision/recall, AUC in „human language”,
• ML with the help of no-code tools:
  
  • using AI for feature design, interpretation of model results, generation of descriptions and visualizations,
• ML models vs. LLMs in an analyst's work:
  
  • when to reach for a classic predictive model, and when it is enough,
  • an LLM with a good prompt or RAG.

7. Creating your own tools with the help of AI

AI can help analysts turn a clearly described problem into a usable tool without writing code. The section focuses on preparing prompts that lead to templates, calculators, checklists, simple automations, macros, and queries, along with practical rules for building reliable in-house solutions.

• how to describe a problem so that AI can design a tool,
• file templates, calculators, checklists,
• simple automations, macros, queries,
• best practices when building „your own tools”.