Data-Hub Sholudchenko: Streamlining Data Labeling and LLM Fine-Tuning
![© Data-Hub Sholudchenko](https://www.trendingtopics.eu/wp-content/uploads/2024/07/Untitled-design-36-780x439.jpg)
Yevgen Sholudchenko founded „Data – Hub Sholudchenko“ just over two years ago. The entrepreneur and former data science consultant was born in Kiev, grew up in Germany, lived in the Philippines and is now based in Vienna. Data-Hub Sholudchenko provides comprehensive data labeling and annotation services, as well as manual quality assurance for ML/AI models. The DataScience Startup primarily focuses on providing manpower and expertise for data labeling. Additionally, a growing business area involves fine-tuning custom LLMs using feedback from specialists like doctors and engineers.
Startup Interviewer: Could you start by telling me a bit about yourself and your background?
Yevgen Sholudchenko: I originally started as a one-man data science consultant, and in some of my projects, I needed data labeling and conducted a few experiments in countries like India. However, I ended up re-labeling all the data myself due to the low quality. Therefore, I thought maybe I should build my own team.
Who is on the founding team?
Originally, it was just me. Later, my partner Enzell, based in the Philippines, joined.
What is the story behind your startup? How and why did you get started?
Since I had previously lived in the Philippines, I had quite a few good friends there. I called a friend and told him I might have a project for him. Originally, we just started with him and his two siblings, but then we grew. Today, we provide all types of data labeling and annotation services, as well as manual quality assurance of ML/AI models. We also work on domain projects where you might need psychologists, doctors, or engineers who can interact with custom LLMs and provide high-quality feedback on their performance.
We have ongoing projects where we provide constant support and other projects where we work for 2-3 months, then take a break so the team can retrain the model and continue again. During peak times, our team consists of around 45 people in the Philippines.
How does your startup stand out from the competition?
Our USP is that before every project, we spend a lot of time understanding our customers. Even a simple task like ‚make segmentations on the buildings‘ can be understood and executed in many different ways. For example, some customers don’t require very precise segmentation and a basic bounding box might be sufficient, while others might need segmentation almost at the pixel level. To understand these needs, we talk extensively with our clients and also conduct a custom onboarding for the team. The team does a few iterations, and only if all team members meet the desired quality thresholds set by the customer, do we start the actual project.
Another focus area for us is that often the customer has a lot of data but may not realize all the potential insights it holds. They might ask us to label based on categories A, B, and C. However, if we know the actual final goal, we sometimes discover that the data might also contain categories D and E, which could also contribute positively to the client’s objective. So we don’t just blindly label; we always strive to generate even more value than originally anticipated.
What technologies do you use, or what in-house tech have you developed?
We are not a software provider; instead, we primarily provide the manpower and expertise for labeling. We have already worked with many different tools, including open source, custom-built, and others.
Can you describe your typical customers and how do you reach them ?
Our target group are AI and data science companies. Occasionally, we also work with universities on research projects, but this is rather rare because they usually require funding, and it’s a very tedious process. We do the good old cold acquisition through different social media channels such as LinkedIn, etc.
What about previous financing? Are there already investors?
We are bootstrapping, and so far, we do not need any investors since we are already self-sufficient and can pay our bills. However, we have made a few strategic partnerships where we offer arrangements like royalty deals or commissions for our partners who bring us clients.
Could you explain your business model? How does your startup generate revenue?
That is quite simple: You either have a certain number of images or text that needs labeling, or you have an LLM that you would like to fine-tune further. We provide the people who can do it. We basically charge an hourly rate; however, we always try to provide a benchmark of how much data we can process. For example, for some projects, we can label 60 images per hour, while for others, it’s only 5. This is always discussed in great detail with the customer – we try to manage all expectations early on.
What are the next steps for your startup? Do you have specific goals for the future?
Right now, we are just trying to keep expanding and growing. We have already had quite a few international clients – in addition to clients in Austria, we have also had some in Switzerland, Denmark, Norway, South Korea, the USA, and a few other countries.
Could you share any advice or lessons learned that might be helpful for other aspiring entrepreneurs or startup founders?
Don’t overthink it – just do it. If it fails, analyze what went wrong and try something new. Nobody really knows what they are doing.