Antoni Czolgowski

Meet Antoni:

From rural libraries to open-source AI research, Antoni Czolgowski's first year with the Master in Data Science program at the University of Colorado Boulder has been defined by one key phrase: Yes. 

While completing his undergraduate degree in Quantitative Methods in Economics and Information Systems from the Warsaw School of Economics, Antoni joined Orange's data engineering team. In 2023, just shortly after ChatGPT launched, Antoni watched an entire organization's relationship with AI transform in real time. 

Now at Boulder, his foundation became a launchpad: independent research, a supercomputing allocation, a published R library on CRAN, an ultramarathon and a hackathon win later, he's closing in on a degree that looks a lot like the versatile, technically deep career he set out to build.

You started at Warsaw School of Economics studying economics broadly — what pulled you toward the quantitative and data side of the field?

My interest started during undergrad, when I discovered big data technologies. It was interesting to think ‘how do we extract big data sets?’ but since then, everything has changed so fast. I have experienced the first two years of the AI revolution both as a student, and an employee within the data science field. Especially during my internship with Orange. 

ChatGPT was released in November 2022, and when I joined the team in 2023, no one had heard of AI. I experienced the change first-hand when management first learned of AI-use and adapted their approach. I’ve seen a full span of approaches with AI – from limited use in the beginning, to a full adoption by the time I was preparing to leave.

I realized then that being up to date with new technologies, eager and bold, gave me an advantage. And I’m still like that. I’m optimistic about artificial intelligence. I've spent time with OpenClaw, an open-source framework for self-hosting personal AI agents on your own machine, to run that kind of infrastructure end-to-end. Now day-to-day the Claude ecosystem is enough: Claude Code in the terminal and the desktop app cover what I need. 

Your research touches on something a lot of people in AI are starting to talk about now: cultural alignment. How did you first get interested in that area, and how did the project take shape?

My research originally began while building my thesis in my undergraduate degree. My research quickly became ambitious and I started working closely with Doctor Magdalena Smyk-Szymanska, who I had previously taken a class from. The work was about using econometric tools to measure real-world social patterns. At Warsaw I came at it from the econometric side. I had the statistical toolkit, and I went looking for problems where it could apply. The biggest shift since then is that I now start from the problem instead. Cultural alignment in LLMs is what I want to measure, and I pull in whatever tools actually fit, including LoRA fine-tuning, distributed inference, and methods I had to learn from scratch. That reframe is honestly the growth I'm most proud of in the past year. 

The field of cultural alignment in the AI field is more and more important. I compare three open-source models, Gemma 3 (US), Bielik (Poland), and Qwen 3 (China), against real human responses from the World Values Survey, then apply LoRA fine-tuning to the demographic profiles where the models perform worst. Commercial models like ChatGPT, Gemini, and Claude dominate everyday use, but the open-source ones were a deliberate methodological choice: with access to the weights, you can measure bias and actually try to reduce it. Most of the workflow runs in R through PacketLLM, my own library, published on CRAN. 

The idea came from my own experience as an early adopter. Treating LLMs as advisors, I noticed the answers didn't reflect the cultural context I was operating in, even after the model had collected a lot of information about me. So the questions followed naturally: do model creators implant their own worldview into the model? Is it the training data, or the lack of diversity in it? And which demographic groups do models understand the least? 

So your research was already in motion before you arrived at Boulder. How did you find a way to continue it within the professional MS-DS program structure, where research is not traditionally required?

I reached out to one of my professors about my desire to continue my research, and was led to independent research. It was very natural. Independent Study at CU is essentially a one-on-one research arrangement: you pair with a faculty advisor, define a research question, and the work counts as graduate credit. 

For me, this meant I could build my cultural alignment research directly into the MS-DS curriculum, which made it financially feasible to pursue in depth and spread the workload more evenly across the program. 

I’m pleased to say the arrangement ultimately produced a paper accepted at OSSConf 2026 in Žilina, where I’ll be presenting my work!

You gained access to CU Boulder's Alpine supercomputing cluster through your independent study. For students who don't know that resource exists — what has it actually made possible for your work? 

If you google, you can find many CU Boulder supercomputing opportunities, like Alpine. My basic access didn’t really provide much support, but having the independent study as an official course helped me gain access to more in-depth research support. 

I can store my models on the supercomputer and run queries in parallel, resulting in faster outcomes. My queries can also run for 24 hours, compared to the one hour provided under the basic CU Student access. 

I believe more MS-DS students could take advantage of Alpine, as it could allow them to complete in-depth review more quickly while also building their working knowledge of supercomputing. 

That’s great to hear. Are there any more resources at CU Boulder that have shaped your work in ways you didn't anticipate when you arrived?

Yes! Before the school year started, our academic advisor Mariah sent us an email for an Interactive Case Competition. I organized a group of my classmates to compete in a competition to assess the needs of underrepresented communities as it regards broadband coverage, resulting in potential strategies to create additional access. 

We became friends as we worked together, attempting to solve real problems with data, but for people. We currently have the best tools in human history at our disposal, and I believe with them we can make a real impact. I’ve always believed that my role as a data scientist can have a societal impact, but this competition further cemented that belief.

Our first semester’s courses proved useful -- especially our statistics and data structures courses. With that knowledge, we created comprehensive modeling using online public data to understand what drives the broadband coverage and where underrepresented communities reside in Colorado. We met with local broadband providers within Park County to understand their specific broadband constraints. We talked to community members in their local grocery stores and libraries, which also happen to operate as local community centers. 

By collaborating with public libraries across underrepresented communities, computers could be equipped with tools to further community tool access and computer literacy. And as community members become more familiar with digital tools, they’re more likely to voice desires to gain their own broadband connections. 

This all happened in the first semester. I was thinking, “Do I have time for this?” But I chose to say yes. I competed with the best schools across the US. I met some of my best friends. We even won a small award. It taught me not to say no. And it’s something I’ll bring with me even if I decide to pursue my PhD. 

How have you gotten involved in the broader data science community at CU Boulder?

I've gotten involved through various competitions, like the Spectrum Broadband Interactive Case Competition, but also through the Data Science Student Association (DaSSA) and the startup scene on campus through Startups2Students.

For the DaSSA hackathon this year, I reached out to a fellow student who'd presented strong work, asking if he wanted to team up. We invited some friends, built a team, and won. Our solution tackled advertising niche products on Reddit.

Startups2Students has been just as formative. One conversation there turned into a data scientist offer at a Denver-area startup for summer 2026. I never sent a CV. Between the labs, the student associations, and the startup events, you could be at a DS-adjacent event every week here. 

It sounds like year one has been a whirlwind! What does year two look like for you, Antoni?

Thanks to this masters degree, I feel strong on the technical side. I feel I can do a lot of things at once, especially with the help of AI agents, and now I’m eager to explore the hardware side. 

This master’s program allows me to request electives that directly align with my own career goals, so I've requested an elective in robotics. The bigger goal is to grow into a versatile tech entrepreneur: someone who can always match the right tools to the idea in front of him. I feel strong on the software side. I want to feel the same way about hardware, and robotics is the first gap I'm closing. 

Is there any advice you would give to someone who is where you were two years ago?

My advice would be to speak. Treat your two years as the only chance in your life to be in such a nice environment. Even if it doesn’t work out, asking is better than regretting. Everyone has ideas – but without communicating them, you can’t realize them. 

Say yes. And more than saying yes, create the opportunity for yourself when you need to. Make your connection. The tools will show up. 

When asked what drew Antoni to Boulder specifically, his answer was simple. The Denver International Airport (DIA). Having such straightforward navigation to his home country eventually sealed the deal. But airports don't write 60-page papers or hold a 4.0 – and that's what his first year here has actually been. He’s been dedicated to the professional masters grind, but isn’t afraid to be active in the local hiking and ultramarathon communities. 

"The Pro Program made the academic grind possible. The Boulder environment made the first ultramarathon possible."

From the MS-DS program at CU Boulder, we hope to see Antoni summiting more mountains, breaking new personal marathon records, and advancing additional data science research in the years to come. 

  Are you an MS-DS student? Email datascience@colorado.edu for a special student resource created specifically by Antoni and learn how to take advantage of CU Boulder's Alpine Supercomputer resource.