Michał Podlewski: Some time ago, we spoke with mathematician Dr. Bartosz Naskręcki about how AI is changing mathematics. There is a widespread belief that biology may be the next field to undergo major changes. From your perspective, is AI already significantly changing the way work is done in this field, or is that still a thing of the future?

Joanna Krawczyk: Let's start with something that might seem obvious - biology is probably the most complex system we are trying to model with AI. From diagnostic tools to protein structure prediction to understanding cancer, it poses thousands of questions we are still trying to answer. At Ingenix, we look at AI in biology primarily through the lens of discovering and developing new therapies.

I believe we are at a genuinely fascinating moment - one full of both real progress and deep contradictions. AI is advancing at an extraordinary pace and has already transformed parts of biology. AlphaFold is perhaps the clearest example: it showed that machine learning could solve a problem that challenged structural biologists for decades.

But if we ask whether this progress is translating into better medicines, the picture becomes much more complicated. Despite steadily increasing R&D investment, the number of new therapies approved per billion dollars spent has been declining for years. We have more biological data than ever before, sequencing is cheaper than ever, and computing power continues to grow, yet we are not seeing a proportional increase in novel therapies reaching patients.

I don’t see this as a failure of AI. Rather, it highlights the limits of today’s approaches. AI performs best when problems are well defined and supported by abundant, high-quality data. That’s exactly what made AlphaFold possible: decades of high-quality experimental protein structures provided an exceptional training signal. Drug discovery is a fundamentally different challenge. That’s why, despite remarkable progress, I think we are still at the beginning of AI’s real impact on therapeutic discovery.

The founders of Ingenix, Piotr Surma and Adam Dancewicz, are experienced entrepreneurs who already have a successful AI startup under their belts. However, applying that experience to the complex world of biology is a challenge on an entirely different scale. What made you believe they would succeed this time as well, and what led you to join the team?

As a bioinformatician with a mathematical background, I always wanted to work at a place where I could combine my passion for biological data analysis with building machine learning models. For a long time, there simply wasn’t a place like that in Poland, so I was genuinely excited when I first heard about Ingenix.

Before joining the team, I was rather skeptical about buzzwords like “simulating clinical trials,” “virtual cells,” or “virtual twins.” My previous experience at a biotech company had shown me how easily these ambitious visions can run into practical challenges with biological data. What convinced me during my conversation with Piotr and Adam was their honesty. They said openly: “We don’t believe we’re going to solve all of biology, collect all the data, or build a single model that does everything.” But they were convinced that by bringing together an interdisciplinary team, we could make real, steady progress toward that goal. To me, that was far more convincing than another promise of a revolutionary breakthrough.

In my view, Ingenix stands on two strong pillars. The first is experienced founders who provide us with the best environment to work and grow. The second is an exceptional team of researchers, engineers, and bioinformaticians. It may sound cliché, but it’s genuinely a pleasure to work in a place like this. We’re a close-knit group of people who truly believe in the company’s mission and push each other to do better every day.

Ingenix is a startup that aims to change the way we use AI in biology and drug development. In your view, building a single large model will not bring about a breakthrough in these fields. Why do you think this is a dead end, and what, then, is your alternative?

This comes down to the nature of biology itself. Let’s start with the fact that biological systems are dynamic and nonlinear - small changes in inputs can lead to complex changes in outputs. A single point mutation, for instance, the substitution of one nucleotide at a so-called hotspot, can set off a chain of events that leads to cancer.

On top of that, biology is multi-scale. We can observe it at the level of DNA, RNA, proteins, single cells, tissues, or entire organisms. Each layer reveals a different slice of reality, and no single measurement gives you the full picture. To understand what’s actually happening inside a cell, you have to integrate information across multiple modalities at once.

And then there’s the most practical issue of all: biological data is far from ideal. Measurement technologies have real limitations - the data is noisy and highly variable across experiments. On top of that, biology is a field where the data tends to be “wide rather than deep”: we often measure the expression of tens of thousands of genes across only a few hundred samples, which poses a real challenge for classical machine learning methods.

Taken together, these challenges mean that biology requires very different AI architectures from those used in domains like text or images. The problem isn’t that today’s models are too small. It’s that we don’t yet have architectures designed for biology in the first place.

That’s why we don’t think the breakthrough will come from one single foundational model trying to learn everything at once. Models like that are great at reproducing biology they’ve already seen - but they struggle with what matters most to us: generalizing to new patients, combinations of mutations, or disease mechanisms.

At Ingenix, we’re going the opposite direction. We build specialized models, each tailored to a specific scale and data type, and then connect them into a Biological Reasoning Engine that can reason across all of those levels at once. I like to compare it to an interdisciplinary team of scientists - no single person knows all of biology, but a team that works well together can see far more than any one person could alone.

Over the past year, you’ve been working on a completely new AI architecture. What have you managed to build during that time?

Over the past year, we have been working on the first version of the Biological Reasoning Engine, which meant advancing on several fronts at once.

On the modeling side, we built specialized components to handle different types of data - from transcriptomic and molecular data to models predicting the effects of genetic perturbations or drug treatments. We treated each of these modalities as a distinct research problem, rather than simply reducing them to text.

Alongside that, we developed a reasoning layer and an agent system that draws on knowledge from the literature, patents, clinical trials, and quantitative experimental data, translating it into research hypotheses.

The first result that particularly excites us is early signs of zero-shot generalization in our models - the ability to reason in new biological contexts the models had not previously encountered. The second is the outcome of our latest internal hackathon, where we built a prototype of an autonomous, self-evolving bio-researcher. The question was straightforward: can the agent enter a completely new biological domain, given only a handful of examples from experts, and quickly learn to work effectively within it? The models developed during this project clearly outperformed publicly available solutions, both on external benchmarks and in our internal tests. This is the first strong evidence that it is possible to build systems that do not merely memorize biological patterns, but adapt to new biological contexts.

We describe some of our work in publications – if you're interested in the technical details, you can find them at https://ingenix.ai/evidence/, along with our blog posts:

Ingenix.ai

By Adam Dancewicz

In your press release, you wrote that the findings—which took the biotech company several years of research and cost millions of euros to develop—were generated by your AI in just a few minutes. Can you share more details on how that happened?

This was a very important moment for us because it was the first time we had the opportunity to test the Biological Reasoning Engine on a real-world research problem.

A biotech company came to us with a specific oncology question, a problem they had been investigating for several years. Our system generated a set of hypotheses based on the analysis of multiple biological modalities at once. It combined biomarker analyses across cell-line and patient-derived data with in vitro evidence, clinical results, and data spanning the molecular, cellular, and population levels. Collecting data, running the analyses, interpreting the results, and building the models - this process usually takes months. Our system performs the analysis within tens of minutes.

However, the real challenge is what happens next: synthesizing and prioritizing the hypotheses. Individual analyses can produce dozens of potential signals - the hard part is identifying which ones have the strongest biological rationale and are most worth pursuing.

That’s why validation was so important to us. The hypotheses that our system generated largely aligned with the directions the company had arrived at after years of its own research, some of which they had already confirmed experimentally. This gave us confidence that the Biological Reasoning Engine could significantly accelerate their research process and help them make key scientific decisions.

This does not mean that AI replaces biological experiments. However, it does show that it can narrow down the set of the most promising hypotheses in a very short time and help scientists make better decisions about which hypotheses are worth testing in the lab. As a result, even a small biotech company can iterate on not just a few hypotheses, but dozens of promising directions simultaneously.

Your flagship solution is the “Biological Reasoning Engine”. How does your AI “understand” biology? What does it take for artificial intelligence to truly understand biology?

I’d actually be careful with the word “understand” - it depends on what we mean by it. A model doesn’t understand biology simply because it has read millions of papers. To me, understanding means being able to connect different biological layers and explain how molecular events translate into cellular behavior and ultimately affect a disease process.

Most AI systems today are excellent at analyzing individual types of data: text, images, or biological sequences. However, when multiple models are combined, they are usually connected only at the level of their final outputs. Biology doesn’t work that way. The same biological process leaves signals across many levels at once: in the genome, gene expression, protein activity, cellular state, and ultimately in the patient’s condition. If we want to understand what is happening, we cannot analyze these layers independently.

Let’s take a concrete example: what happens if we give a specific compound to a patient with a particular cancer mutation? The answer does not come from the molecule’s structure, gene expression profile, or literature alone. The system needs to reason across all of these dimensions simultaneously, combining information about the compound, the state of the cell, active signaling pathways, and available experimental evidence, to build a mechanistic hypothesis explaining why we would expect a particular response.

This is why our approach does not combine models only at the level of their outputs. We integrate them earlier, at the level of their internal representations. Each specialist model remains an expert in its own domain, but reasoning happens across a shared map of relationships between biological modalities. For us, this learned map, the connections between different layers of biology, is the most valuable part of the system.

Transparency is equally important. From the beginning, we wanted to avoid building another black-box system that simply returns an answer. The system should make its reasoning transparent: which evidence it used, what biological mechanism it proposes, and where the limits of that hypothesis are.

Does that mean AI “understands” biology? I would be cautious about making that claim. I would say instead that it can reason across different biological scales and build coherent mechanistic explanations in a way that was previously extremely difficult, because no individual scientist can simultaneously integrate across so many sources of information and biological scales.

The internet has been the fuel for LLMs. What is the equivalent of the internet for biological AI? Laboratory data? Simulations? Experiments?

The shortest answer is: experimental data. But this is where the fundamental difference between biological AI and language models begins. The internet is vast, cheap, and easily accessible. Biological data is almost the exact opposite - every measurement requires an experiment, and experiments are expensive, time-consuming, and often affected by technical noise.

But the challenge is not simply having more data. It is having the right kind of data. For biological AI, the structure of the information matters enormously. We need heterogeneous data that capture causal relationships across processes and scales. A good example is perturbation experiments, where we deliberately change a biological system, for example, by treating a cell line with a drug or switching off a specific gene using CRISPR, and observe the consequences. There is a simple engineer’s principle behind this: if you want to understand how a system works, try to break it. Perturbations allow models to move beyond correlations and start learning causal relationships. Even more valuable are multimodal datasets that connect different biological scales within the same experiment: perturbations across different cell types and contexts, paired patient data before and after treatment, or multiomic profiles linked to therapeutic response. Today, we have very few publicly available datasets of this kind - and that is one of the biggest bottlenecks for progress.

Without better data, we risk falling into two intertwined traps. First, models become very good at reproducing the biology we already know. Second, the benchmarks that drive progress in the field often reward familiarity over discovery. Many public evaluation datasets focus on well-studied biological contexts, measuring how well models reproduce existing knowledge rather than their ability to uncover new biological mechanisms.

Do we really need dramatically more data? Not necessarily. The more important question is: what data is actually valuable for teaching models to reason about biology? We believe we need two things at the same time: more heterogeneous, causal datasets that capture how biological systems work, and rigorous benchmarks built through a collective effort across the community.

What does your day-to-day work with AI look like? In what tasks do you find AI indispensable, and where do you think it still can’t be trusted?

At Ingenix, AI is primarily a tool that accelerates our everyday work. I have to admit that the progress I’ve seen over the past year has been remarkable - especially when it comes to AI-assisted coding. Models now help me prototype ideas faster, explore the literature, and monitor computational experiments. As a result, I can spend more time on the actual research and less on repetitive tasks.

At the same time, there are areas where I remain deliberately cautious: especially data exploration, data preparation, and interpreting model outputs. In biology, the quality of the data we use to train and validate models is absolutely fundamental. If the underlying assumptions are flawed or the data quality is poor, even the most advanced model will simply produce more sophisticated errors.

The same applies to interpreting results. In machine learning, it is surprisingly easy to achieve impressive performance for the wrong reasons. The metrics may all look excellent: the loss is low, accuracy is high, and yet, when you look deeper, you discover that the model has learned a technical artifact rather than a real biological signal. This is where human judgment is still essential. You need to take a look at the results and ask the most important question: does this actually make biological sense?

Where do you think we’ll be in 5–10 years thanks to systems like yours? Will we reach a point where AI is not just a tool for researchers, but the primary author of discoveries? And what does that even mean for science as a human institution?

I’m not a big fan of speculating about the future - biology has taught me humility too many times for that. But I believe that at Ingenix, we have a vision worth fighting for.

Today, it takes more than a decade, on average, to go from identifying a promising drug candidate to clinical approval. That’s a decade during which patients are waiting for therapies that could potentially help them. I’d like to see AI shorten that timeline over the next 5–10 years, not by replacing scientists, but by accelerating every stage of the development process: generating and prioritizing hypotheses faster, designing better clinical trials, and detecting signals earlier that today get lost in the noise. That is precisely Ingenix’s mission: shortening the path from discovery to life-saving therapies.

As for whether AI will become the primary author of scientific discoveries, honestly, I don’t know. But I do know that if systems like ours help scientists ask better questions and test them faster, that will be a win for both science and patients. Tools can never replace curiosity, and curiosity is what drives discovery.

***

Joanna Krawczyk – Senior Bioinformatics Data Scientist at the Warsaw-based startup Ingenix, with professional experience spanning both academia and industry. She previously worked in Ewa Szczurek's interdisciplinary Computational Medicine laboratory at the University of Warsaw, as well as at the biotech company Ryvu Therapeutics.

***

Links:

1. LinkedIn Profile

2. OpenReview Profile

3. Ingenix Substack

4. Ingenix website

Joanna Krawczyk with Frederic Grabowski at ICML 2026 in Seoul, South Korea presenting two papers on multimodal language models for omics and molecular reasoning:
🧬 OmicsLM: A Multimodal Large Language Model for Multi-Sample Omics Reasoning
🧪 Bolek: A Multimodal Language Model for Molecular Reasoning

Early Bird Registration for ML in PL Conference 2026 is open! The conference will take place on 8–10 October 2026 and bring together the machine learning and AI community from across Poland and beyond.

Selected applicants will receive access to a discounted conference ticket.

Register at the link below.

ML in PL 2026

On May 8–9, Poznań hosts GHOST Day: Applied Machine Learning Conference. This isn’t a high-level “strategy” summit; it’s a gathering for the people who spend their days building real-world ML systems. No corporate fluff. Just math, code, and breakthroughs.

It’s a rare opportunity to see what research teams from the world’s top AI institutions are actually working on.

The Lineup: Featured Speakers & Topics

• Michal Valko | Isara Labs

“The Genesis of Predictive Architectures: How AI Can Extract Meaning from Unlabeled Data”

• Przemysław Biecek | Centre for Credible AI

“Decode AI: A Few Thoughts on How to Analyse Complex Models (and Why)”

• Ondřej Dušek | Charles University in Prague

“How (Not) to Find Errors in LLM Outputs?”

• Pierre Ménard | Meta

“ARE: Scaling Up Agent Environments and Evaluations”

• Sagar Vaze | @MistralAI

“What’s Next in Vision Language Models?”

• Rebekka Burkholz | CISPA Helmholtz Center for Information Security

“Towards AI that is Smart and Sparse”

• Marina Esteban-Medina | ETH AI Center

“Learning Drug Responses Across Domains for Precision Oncology”

• Ivan Cimrak | University of Zilina

“If Mammography AI Works, Why Isn’t It Everywhere?”

• Karolina Drożdż & Kacper Dudzic | IDEAS Research Institute

“Patterns vs. Patients: Evaluating LLMs against Mental Health Professionals on Personality Disorder Diagnosis”

The roster also includes dr. Bartosz Naskręcki, Witold Wydmański , Krzysztof Ociepa, prof. Artur Dubrawski (Carnegie Mellon), and Mateusz Olko.

Why It’s Worth the Trip

• Real Recruitment: Sponsor booths from Allegro, AMD, Datarabbit, Google, and Pearson aren’t just there for branding. They are actively looking for talent. Bring your CV.

• High-Signal Networking: After the first day, we’re heading into central Poznań for a networking event. It’s the best place to debate architectures with people who actually know what a transformer is.

• Community Driven: The event is organized by the incredible team at Poznań University of Technology

Deep Dive: The Interviews

If you want to hear more about the vision behind the conference, check out the full interview with the GHOST Day team:

• English: fast-takeoff.com/p/ghost-day-2026

• Polish: trajektorie.pl/p/ghost-day-2026-wywiad

When: May 8–9, 2026

Where: Poznań University of Technology, Poland

Tickets: ghostday.pl/#tickets

See you in Poznań!

For decades, mathematicians used a whole arsenal of functions—logarithms, powers, trigonometry—without realizing that they could all be reduced to a single, universal operator. Polish astrophysicist and mathematician Dr. Andrzej Odrzywołek (Jagiellonian University, Kraków) showed how to do this, and his work sparked widespread attention. In an interview for fast-takeoff.com, we discuss mathematics that looks like science fiction, whether analog computers can make a comeback, and how artificial intelligence is changing the way science is done.

***

Your latest paper, “All Elementary Functions from a Single Operator”1 has caused quite a stir online. But if exponential functions, logarithms, and their connections have been known since Euler, why has such a simple, universal operator eluded mathematicians for centuries?

Dr. Andrzej Odrzywołek: Because the scientific procedure may not be clear to the broader community, let’s clarify: this is not a [scientific] paper but a preprint. The results have not yet been formally peer-reviewed and finalized.

It is common knowledge among experts that only a few operations are sufficient to provide basic “elementary” functions, which I define as those known from scientific calculators. For example, Mathematica, used for calculations by theoretical physicists, requires only four operations: addition, multiplication, exponentiation, and the base logarithm. Converting multiplication to addition and vice versa is also, and may have been obvious before calculators appeared. Every schoolchild, student, or engineer had to master the use of tables or a slide rule. The exponential function exp(x), its inverse (the natural logarithm ln(x)), and the four operations (+, −, ×, ÷). That’s enough. Further steps from the recommendation (notice of centuries to exaggeration) are simply not followed and are not implemented. The key formula for the logarithm in EML form requires 7 tools and three applications, which is borderline “natural.” EML itself requires 3 operations on a 36-key calculator, a number that can be checked with a naive algorithm up to about 6 million (2⁷ × 36³). Too many, due to the concatenation at the end, but few enough that someone had to do the math in the first place.

Some people admire the elegance of EML, others ask, “OK, but why?” How do you respond to the accusation that it’s ultimately just mathematical gymnastics, a kind of “art for art’s sake”?

Three questions in one :-)

The concept of elegance is subjective. It’s more accurate to talk about complexity or simplicity. One binary operation and the number 1 are less than a set composed of several functions of a single variable, binary operations, and constants. Formulas compiled into EML can seem ugly, like the pi formula circulating online. I’d call it rather lengthy. However, the EML operator itself is indeed elegant; its form really surprised me. I expected something worse, maybe not necessarily as strange as, say, Minkowski’s ?(x), but rather among special functions rather than school functions.

I needed a single operator for what’s called symbolic regression, i.e., searching for mathematical patterns in data sets. Recently, such techniques have been sidelined, because AI does the same thing much better. The main problem in SR is choosing a base: sine or cosine, tangent or cotangent, addition or subtraction, etc. Using too many operations complicates the search; too few, and we risk missing some valid expressions altogether. For a long time, I’ve had the idea for an automatic verifier that, before running a search, for example, in PySR, checks whether the set of constants (e.g., 2, pi, e), functions (e.g., sin, sinh, cos, cosh), and operations (e.g., +, −, ×) used will actually find any expression. While testing the verifier (VerifyBaseSet), I fed it various things and observed what happened. Then I realized that with just one operator, the problem disappears altogether.

Without a doubt, EML is a kind of intellectual gymnastics. But as Stanisław Ulam wrote:

In your publication, you also write about a prototype EML compiler and the potential for analog computer design. Does this mean that in the future, our processors could operate on a completely different, simplified architecture? Where exactly might EML find the quickest physical application?

Analog computers were long ago replaced by digital ones, but that doesn’t mean they’ve ceased to exist. Their use is limited by technical problems, one of which is the inability to generate arbitrary elementary functions. EML could fill this gap. Generating an exponent is simple: it’s the law of capacitor discharge; subtraction is performed by an operational amplifier. But along the way, we need complex numbers. The path from theory to practice won’t be straightforward. The brain undoubtedly operates in analog mode, and we can’t even replicate the brain of a fruit fly.

Just three years ago, the world of mathematics was captivated by the “einstein” tile: a single prototile that by itself forms an aperiodic set of prototiles. In a sense, you yourself have a similar invention to your name: a single operand that can express any real-world function. Do you think other fields are still hiding their “universal building blocks” from us?

These are related. I myself invented and hold a patent for a three-dimensional polyhedron whose shape allows you to build an igloo. You can find a description in the journal Eureka (”How to Build the Perfect Igloo”2 2014), published by mathematics students at Cambridge, and its model is held by the NKF UJ. I believe that redirecting the computing power of hundreds of thousands of GPUs currently used to train AI to search the solution space of various problems could reveal many surprises.

Apparently, EML is just the beginning, and you’re already exploring its ternary variant, which doesn’t need the constant “1” at all. Why is eliminating this one extra digit so important to you, and what would it take to find a fully self-sufficient operator?

The need to consider the possibility of a specific constant at every level complicates calculations and increases the number of parameters in the EML tree. It also acts as a kind of discrete switch that removes the logarithm from EML. A model without any specific constant, containing only variables x, y, z... would be fully continuous. Such a binary operator would be even more similar to NAND. I haven’t been able to find it. Perhaps it doesn’t exist or is not an elementary function.

At the end of the publication, you openly admit that modern language models have aided you in your work, including translating code into Rust. What’s your day-to-day collaboration with AI like?

Language models have accelerated many things. Firstly, effective translations and multilingual search. Now, on X, I immediately see, for example, posts originally in Japanese that would probably be completely incomprehensible to me without them. From a circuit diagram alone, I wouldn’t have guessed that someone was presenting a proposal for a hardware EML implementation. Secondly, a literature review, including historical texts in Latin or French (Cotes, Liouville). This is very helpful, especially for interdisciplinary research in an era of narrow specialization. In a sense, LLMs fulfill the promise that Google once gave us and then took away: the ability to find any information we need.

Another issue is language proofreading. However, this mainly works with standard texts and requires extreme caution, as the problem of hallucinations still persists. Code translation is the latest AI achievement; a breakthrough occurred in December 2025. Currently, the problem isn’t writing the program itself, but rather clearly defining what it should do and how. It turns out that the best way to explain how a program should work is to have another program do the same thing. An AI agent like GPT Codex or Claude Code can independently run and test programs, allowing for efficient code translation, for example, from Mathematica to Rust. Having two versions allows for easy verification of whether they function identically. The difference is speed: the original VerifyBaseSet requires 40 minutes, but after rewriting it in Rust, it takes just seconds. AI also facilitates working with different operating systems; the repository included with the preprint works on Windows, Linux, and MacOS.

***

Dr. Andrzej Odrzywołek – Polish astrophysicist and mathematician, member of the Department of General Relativity and Astrophysics at the Institute of Theoretical Physics of the Jagiellonian University in Kraków. Author of the paper “All Elementary Functions from a Single Operator,” which caused a stir in the international scientific community. Holder of a patent for a three-dimensional polyhedron for igloo construction (the model is in the Jagiellonian University collection).

Links:

1. UJ employee page

2. LinkedIn

3. X/Twitter

In an era where "AI" has become a buzzword often stripped of its technical weight, GHOST Day: Applied Machine Learning Conference stands as a beacon for those who value engineering depth. What started as a "horribly optimistic" initiative by students at Poznan University of Technology has evolved into a premier European event, attracting researchers from ETH Zürich, Meta, and Mistral AI. This is not just another tech talk—it is a space where scientific rigor meets real-world implementation. We sat down with the organizers to discuss the evolution of the "GHOST" spirit and why 2026 is the year to move past the hype and focus on what truly works.

***

What exactly is GHOST DAY, and what is the story behind your intriguing name?

Piotr Wyrwiński (GHOST’s Supervisor): GHOST stands for Group of Horribly Optimistic STatisticians, and the name actually tells our origin story pretty well. It started when Mateusz Lango Ph.D. was teaching Statistics at Poznan University of Technology. A group of ambitious students wanted to go beyond the curriculum, and since Mateusz was doing his PhD in machine learning, he pointed them toward Andrew Ng’s famous ML course on Coursera. That naturally turned into regular meetings, and those meetings grew into what became the GHOST research group. The name stuck because it captures that initial spirit of students who were, let’s say, horribly optimistic about how far a statistics background and sheer enthusiasm could take them in ML. Turns out, pretty far.

Today GHOST is a student organization focused on studying and conducting research in machine learning. Every semester we run study groups and project teams covering different areas of ML. This semester, for example, we have sections on topics ranging from TinyML and quantum computing for ML, through LLMs and robotics, to a team building a flood prediction system using deep learning and satellite imagery. You might have also come across us at hackathons, Kaggle competitions, or other conferences.

GHOST Day: Applied Machine Learning Conference is our flagship event. The idea is to create a friendly, energetic space where ML enthusiasts can exchange experiences and stay current in a field that moves incredibly fast. Our speakers include researchers publishing at top venues like NeurIPS or ICML and experts from companies building real ML-powered products. The conference isn’t tied to any specific programming language or tech stack; it’s about techniques, methods, and algorithms. We also see it as a gateway for talented engineers who want to enter the world of AI, through intermediate-level talks, a poster session where attendees can present their own work, and direct contact with companies in the industry.

The 'Applied' subtitle seems to be a core promise. Is this a clear signal that you focus on real-world implementations and problem-solving rather than just abstract theory?

Szymon Czajkowski (Project Leader): Definitely yes - the name Applied Machine Learning Conference is, for us, a commitment not to remain solely in the realm of theory. While strong academic foundations are extremely important to us (after all, we come from Poznan University of Technology), one of our key goals is to demonstrate how these advanced concepts actually “work” in the real world.

That “Applied” aspect is the core of our community. We want participants to leave GHOST Day with the feeling that AI is not just fascinating theory found in research papers, but above all a practical tool they can use in their own projects. We combine scientific rigor with practical know-how, because we believe that only this combination enables the creation of effective solutions that truly make an impact.

Who is your primary audience? Is the program strictly designed for Computer Science students, or is there room for anyone passionate about the future of tech?

Jakub Bilski (Marketing Team): When we think of “the future of tech” in 2026, our attention most likely turns to the term “artificial intelligence.” The spark that ignited all this momentum was undoubtedly the release of ChatGPT in 2022. This has clear advantages - the AI/ML industry has accelerated at a pace that previous generations of engineers could only dream of for decades. Today, AI permeates almost every field and has become one of the primary directions of technological development.

At the same time, it’s important to keep in mind that with popularity comes a certain degree of superficiality. AI is often reduced to simple, flashy applications that perform well in the media, but do not necessarily reflect the true complexity of the field. As a result, discussions can easily become oversimplified, leading to misconceptions about what artificial intelligence really is - with Machine Learning being a core component of it.

GHOST Day existed long before AI entered the mainstream, and this largely defines the character of our conference. GHOST Day is an engineering- and research-oriented event, aimed primarily at people who already have a certain foundation and want to take the next step. After all, conferences are not about testing your knowledge, but about expanding it.

That’s why we don’t close the door to anyone, and computer science is not the only valid background. Analysts, statisticians, and engineers from other disciplines are equally welcome. However, we do expect a certain level of technological maturity. To avoid any misunderstandings, we are always transparent: GHOST Day is not a course on using commercial AI tools. We will not be teaching how to generate funny videos or how to chat with bots.

This year’s agenda is incredibly diverse—covering everything from LLM error detection and precision oncology to AI in boxing and autonomous waste collection. How did you manage to bridge such different fields into one cohesive program?

Szymon Czajkowski (Project Leader): Our philosophy is simple: we are building an event designed to be a space for every technology enthusiast, regardless of what they are currently working on. We want to unite diverse interests under the banner of AI and ML, creating one shared community.

That is why, alongside medicine or sports, the program includes topics as varied as Self-Supervised Learning—which involves teaching models to draw conclusions without labeled datasets—and neuromorphic systems, which draw inspiration directly from the biological structure of the brain. We discuss model reliability, but also very practical implementations, such as the automated analysis of millions of documents using vision-language models.

It is precisely this diversity that ensures GHOST Day is not just another closed-off conference for a narrow group of specialists. We believe that despite various technical challenges, it is this exchange of experience and shared passion that builds our unique community.

Your lineup includes experts from giants like Meta and Mistral AI, alongside researchers from top-tier institutions like ETH Zürich, University of Edinburgh, and Charles University. How did you manage to bring speakers who regularly publish at NeurIPS and ICML to Poznań?

Wiktor Kamzela (Speakers Coordinator): Each year, we strive to build an agenda that is, above all, as engaging as possible - we look for individuals who not only conduct cutting-edge research but are also eager to share their knowledge and inspire others. The fact that our invited speakers include experts who publish at top-tier conferences is the result of tremendous effort from the entire organizing team.

The greatest recognition of our work is that many speakers are not only willing to return as participants, but also choose to collaborate with us and recommend our event to their peers, who later become speakers themselves. The conference continues to grow year by year, allowing us to invite individuals who, just a few years ago, might not have even considered attending a conference organized by students from Poznań - and who now see us as a compelling alternative to mainstream conferences.

There will be several partner booths on-site. Is this a good opportunity for attendees to bring their CVs and discuss specific job openings or internships?

Adam Dobosz (Sponsors Coordinator): Absolutely! At the sponsor booths, participants will be able to speak directly with company representatives and learn about current job and internship opportunities. This year, exhibitors include (in alphabetical order): Allegro, AMD, Datarabbit, Google, and Pearson.

GHOST Day attracts top students and AI/ML professionals, making it one of the largest concentrations of talent in this field in Poland. Companies are well aware of this and treat the event as a genuine recruitment opportunity, while participants - both those taking their first steps in the industry and experienced specialists - see it as a chance to explore career opportunities.

Conferences are more than just lectures. What are your plans for integrating the community? Will there be networking sessions or social events to talk to the experts face-to-face?

Lidia Wiśniewska (Finance and Logistics Coordinator): Absolutely. We believe that the most valuable ideas are born at the intersection of different perspectives, often in more informal settings. That’s why, after the first day of inspiring talks, we are inviting our participants to a networking event in the center of Poznań. Our goal is to create a space for casual exchange of experiences and for establishing relationships that last for years, and perhaps even evolve into collaborative projects in the future.

Additionally, the agenda for both days includes short coffee and networking breaks. These are the perfect moments to discuss the talks or simply to catch up with others in the industry.

Finally, I would like to ask you to invite the readers of fast-takeoff.com (and beyond) to take part in your event—I personally invite you to attend as well!

Jędrzej Ogrodowski (Marketing Coordinator): We invite you to GHOST Day: AMLC 2026 - a great opportunity to see what research teams from the top institutions mentioned above are working on, connect with others, and further develop your interests. Ahead of us are two days full of inspiring talks, conversations with experts, knowledge sharing, good coffee—and more. The event will take place on May 8–9 at the Lecture Center of Poznan University of Technology. See you there!

Join us: https://ghostday.pl/#tickets

LinkedIn

Twitter/X

Next Frontier AI is the Europe’s most radical initiative designed to discover and define the next paradigm of artificial intelligence.

Managed by SPRIND (Germany’s Federal Agency for Breakthrough Innovation), this initiative provides the funding, freedom, and ecosystem required to establish three new Frontier AI Labs in Europe.

After successful stops in Paris and Munich, Next Frontier AI is landing in Warsaw on April 22. A €125 million funding pool is now accessible to Polish teams:

• 10 selected European teams will each receive €3 million in non-dilutive funding (zero equity taken).

• The top 3 teams can receive an additional €23.5 million over two years.

• The Ultimate Goal: A structured path to raise up to €1B to build a full-scale European Frontier AI lab.

If you are an AI researcher or deep-tech founder working on the next breakthrough—whether it’s radical new architectures, world models, or agentic systems—this is the platform to scale your hypothesis.

The event features a high-caliber lineup of researchers, operators, and investors:

• Carsten Dirks (SPRIND): Next Frontier AI Lead and former COO of Aleph Alpha.

• Prof. Tomasz Trzciński (Warsaw University of Technology): Director of ELLIS Unit Warsaw and NeurIPS 2025 Best Paper recipient.

• Prof. Krzysztof Pyrć (Foundation for Polish Science): Renowned virologist and head of the Jagiellonian University Virology Lab.

• Aleksandra Pedraszewska (vastpoint): Deep tech VC and early leader at ElevenLabs.

• Callum Hill (Creator Fund): Investor specializing in early-stage scientific founders and AI.

Registration is free of charge, and you can sign up at the link below:
https://luma.com/nfai-warsaw?tk=5IRfMv

I hope you will be there.