Introduction
In this episode, host Jacob Edmond sits down with Volodymyr Hresko, co-founder and COO of Coxit, an engineer led AI and machine learning development shop that builds custom AI for document and drawing heavy workflows. Volodymyr walks through Coxit’s path from software development into AI, how the company’s work with Stevens Industries brought them into the millwork industry, and what they have learned from building CaseVBench, their benchmark for testing how well different AI models read architectural and casework drawings. They also get into how shop owners should think about choosing and switching between AI models, protecting sensitive project data, and where machine learning still beats general AI for high volume, repeatable tasks. Whether you are experimenting with AI for takeoffs or just trying to understand what is realistic today, this conversation lays out the tradeoffs clearly.
About Our Guest
Volodymyr Hresko is co-founder and COO of Coxit, an AI and machine learning development company he started with his co-founder Irina seven years ago. He holds a computer science degree and worked as a full stack software engineer before completing a master’s degree in management and finance. Coxit’s work began with C++ projects for radio companies in the United States, moved into Python development, then machine learning, and shifted into AI as the technology matured. Volodymyr and his team have spent the past several years applying that experience to manufacturing, starting with a machine learning project for Stevens Industries before the current AI boom and growing into a company that now spends the majority of its time on AI and machine learning work for manufacturers. He is based in Lisbon, Portugal, while Coxit’s roughly 40 person team is anchored in Lviv, Ukraine, with additional staff working internationally.
What You’ll Learn
- How Coxit’s path from C++ to Python to machine learning to AI led them into the millwork industry through Stevens Industries.
- What CaseVBench is and how Coxit uses it to test AI models against more than 1,400 hand marked objects on architectural drawings.
- Why the newest AI model is not always the best one for reading drawings, and how quickly the rankings can change.
- Why speed and cost per page matter as much as accuracy when picking a model for takeoff work.
- What AI can realistically help with today in a takeoff workflow, and where it still falls short.
- Why Volodymyr recommends building AI tools that are not locked to a single model or provider.
- How shop owners can protect sensitive drawings and pricing data before feeding them into an AI model.
- When it makes sense for a shop to invest in a custom machine learning model instead of relying on general AI.
- Why clients are increasingly coming to Coxit with working prototypes instead of early ideas.
- Volodymyr’s take on which roles in the industry are most likely to change because of AI, and which ones are not.
Where to Learn More
Volodymyr’s LinkedIn: https://www.linkedin.com/in/vovgre/
Coxit Website: https://coxit.co/
https://coxit.co/industry/manufacturing/
CaseVBench: https://casevbench.com/
Final Thoughts
What stands out in this conversation is how fast the ground is shifting under every decision a shop makes about AI. Volodymyr’s biggest point is not which model wins today, it is that shops need to build for change rather than commit to one answer. Between model switching, data protection, and knowing when machine learning actually pays off, the shops that treat AI as a flexible tool rather than a single purchase are the ones set up to keep benefiting from it as the technology keeps moving.
