Is it necessary to train an AI model from scratch?
Almost never. In most cases, it's enough to use an existing model and give it the appropriate context (prompting, RAG, or at most, fine-tuning). Training from scratch requires a huge amount of data and resources that only make sense in very specific situations.
Intelligent Assistants
What type of documents can a RAG system work with?
With almost any text source: PDFs, Word documents, web pages, technical documentation, databases, or notes. What most influences quality is not the format, but how well the content is cleaned and divided before indexing.
Intelligent Assistants
What is a RAG system?
RAG stands for Retrieval-Augmented Generation. Instead of relying solely on what it learned during training, the model first searches for information in a database of documents (manuals, notes, technical documentation, etc.) and uses it to construct its response. This is the technique I used, for example, in my personal chatbot.
Intelligent Assistants
How do I know if it's worth automating something?
My rule: if it's a task I repeat often, it takes up my time, and it follows fairly clear rules, it's a candidate. Even so, it's worth comparing the time you save with the time it will cost to build and maintain the automation; not everything is worth it.
Automations
What types of tasks can be automated?
Any repetitive, rules-based task: sending emails, sorting files, generating documents or reports, processing data, querying APIs, or syncing applications. Many of my lab's experiments started precisely like this, automating something I found tedious to do manually.
Automations
Do you need a lot of data to do data science?
No. The required volume depends on the problem. Many useful analyses come from small but well-prepared datasets; in fact, data cleaning and understanding often matter more than quantity.
Data Science
Does every problem require artificial intelligence?
Not at all. Often, traditional automation or a few well-designed rules solve the problem more cheaply and reliably. AI adds value primarily when it comes to interpreting natural language, documents, or unstructured information.
Artificial intelligence
Is it safe to use AI with private information?
It depends on how it's done. Before pasting sensitive data into a chatbot, it's advisable to check what the provider does with it. For sensitive information, there are alternatives such as accounts with no-training policies, private environments, or models that run locally.
Artificial intelligence
What's the difference between using ChatGPT and building your own AI-powered app?
ChatGPT is a general-purpose tool you use for conversations. When you build your own application, you use the model through its API and connect it with your data, rules, and other tools, so it can work within a process without requiring manual input each time.
Artificial intelligence
Showing 10–18 of 19 questions
✉ SHALL WE TALK?

Let's talk

If we share interests in artificial intelligence, data, or software development, I'd be happy to exchange ideas. You can also contact me for professional matters.

Read the blog