Posted on Aug 30 AI-assisted
Welcome to Best in IT.
I’m Artur Poniedziałek — an IT project manager and technology enthusiast who enjoys turning promising tools into practical, repeatable solutions. This blog is where I share what I learn while working with AI, software, automation and modern IT infrastructure.
I started Best in IT because there is no shortage of exciting technology news, impressive demos or bold claims. What is often harder to find is a clear answer to a more useful question:
Does this actually work in practice, and how can I build or use it myself?
That is the kind of question I want this blog to answer.
What I write about
The main focus of Best in IT is practical AI and technology for people who build, manage or improve real systems.
A large part of my current work and research is around local AI models and private AI assistants. I am interested in running useful models on local hardware, understanding the trade-offs between model size, speed, memory and quality, and finding workflows where AI can genuinely save time without giving up control of sensitive data.
I am especially interested in:
Local LLMs, inference and AI hardware
RAG systems, vector databases, chunking and retrieval
AI agents, orchestration and human approval workflows
Python automation and developer tools
Local image generation with ComfyUI
Speech-to-text, Whisper and video transcription
UI testing with Playwright and vision-language models
IT project management and the practical side of introducing new technology
You will find both explanations for people exploring a topic for the first time and deeper technical posts for readers who want to reproduce an experiment on their own machine.
Why local AI matters to me
Cloud AI is powerful, but it is not always the best answer.
For many teams and individuals, local AI can offer more privacy, predictable costs, lower latency and more control over how data is handled. At the same time, running models locally brings its own challenges: GPU memory limits, driver issues, model formats, hardware compatibility and the gap between a benchmark and a usable daily workflow.
I like testing those details instead of stopping at the marketing page.
For example, I work with local models and automation on Windows and AMD hardware, including an AMD Radeon PRO W7900. My experiments have included using whisper.cpp with Vulkan for transcription, running ComfyUI through its API for automated image generation, testing RAG chunking strategies, and combining Playwright screenshots with local vision models for UI analysis.
The goal is not to claim that one tool is perfect. It is to document what worked, what failed, what needed adjustment and what I would do differently next time.
What I am building and exploring now
Right now, I am particularly focused on building practical AI workflows rather than isolated demos.
That includes private AI assistants with memory, controlled delegation and a clear human approval step; RAG pipelines that can answer questions from business or technical knowledge; and automated content workflows that help turn research and experiments into useful articles.
I am also exploring ways to make local image generation more repeatable. Instead of treating ComfyUI as a manual creative tool only, I am interested in API-based workflows where prompts, models and generated outputs can be tested, versioned and reused.
Some articles will be hands-on tutorials. Others will be benchmarks, comparisons or honest notes from a project that did not go exactly as planned. I think both are valuable. The best technical lessons often come from the part that took longer than expected.
Who this blog is for
Best in IT is for developers, IT professionals, technical managers, consultants and curious builders who want practical context around AI and software tools.
You do not need to be an AI researcher. If you are trying to decide whether a local model is worth running, how to build a more reliable RAG system, how to automate repetitive work with Python, or how to introduce AI responsibly in a business environment, you are in the right place.
My promise is simple: no hype for its own sake, no copied benchmark tables without context, and no “just use AI” advice without discussing implementation.
I will share what I test, explain the decisions behind it, and keep improving the material as tools and models change.
Thanks for being here. If there is a topic you would like me to test or explain, feel free to leave a comment or get in touch.