A smart AI anti-fraud system for an international client, a PoC for a Czech bank with 98% accuracy, and an agent that can perform a company background check in minutes. At Trask, AI isn’t a buzzword. It’s a set of concrete projects on which the team around Anna “Anička” Plašek — IT Project Manager and Business Consultant in the Data Science team — is actively working. How do such projects come to life, what troubles clients the most, and which AI tools save her hours of time?
Anička, we can’t start anywhere else but with artificial intelligence — what do you actually enjoy about it?
What I enjoy most about AI is its dynamism and the fact that it forces me to keep learning. A project I worked on a year ago looks completely different from the projects we’re doing today, and in another year it will likely be somewhere else again.AI isn’t a field where you set something up once and then just repeat it. It’s a constant process of adaptation and innovation.
I’m also fascinated by the speed of development. Every month a new tool or a new approach appears, opening up new possibilities. You have to stay “in the game” and think about how to use it in practice and how to turn it into real value for the client. And I also enjoy that not everyone is fully oriented in AI yet. Some see it as a threat, others as short-lived hype — but when you see real projects and their impact, it all starts to make a completely different kind of sense.
What types of AI projects are you currently working on?
Today we’re working on several types of AI projects. Sometimes a client comes and says: “We want AI in our company.” And they actually have no idea where it could help. Our role is to find the area where the benefit will be the greatest.
Then there are clients who know exactly what they want. Typically these are cases of back office automation — large volumes of manual work, data transcription, document processing, searching in internal systems, or handling requests.
Another strong area is knowledge management. The client wants a tool that can quickly and accurately find relevant information in their internal documentation.
For example, in banking — my colleague Martin Tomis and his team recently handled the deployment of an AI agent for evaluating complaints about card transactions. After adding the “complain” button in the app, the bank was flooded with requests, and the internal team couldn’t process them within deadlines. That’s where an AI agent makes sense: it speeds up decision-making and eases the workload.
Ranking and recommendation projects are also interesting — for example for a foreign discount portal where we optimized the ranking algorithm tailored to the local market.
Speaking specifically… Where else are you deploying AI right now?
We’re currently working on a Risk Agent at one of the largest banks in the Czech Republic. It’s an AI solution built on data in Databricks that helps interpret the outputs of risk models. The AI agent analyses a large volume of data — for example to explain metric changes, identify trends, or uncover reasons for data deviations. It then generates a clear text interpretation of risks for the user. It’s a way for AI to transform complex data outputs into a form that helps risk analysts make quick decisions and conduct deeper analysis.
A very successful project was the Smart Search for another bank. We developed an intelligent RAG platform that replaces manual searching in the internal knowledge database. In the PoC we achieved over 98% accuracy. For an international automotive client we built a similar solution — the goal was at least 80% extraction accuracy, but we ultimately reached 92%.
We’re also working on purely proactive and defensive AI systems. A great example was a PoCin the form of an AI anti-fraud system, where several AI agents together with LLM models process external data, monitor threats in real time, and test rules against new types of fraud.
We also have strong projects at another bank, where we are addressing several practically oriented AI use cases. One of them is complaint categorization, where AI automatically assesses the type and nature of the request so the team doesn’t have to manually review and classify every complaint. The second project is the extraction of notary requests, where AI helps interpret often highly diverse documents, extracts key information, and converts it into structured form. Both projects aim to reduce the high volume of manual work while increasing accuracy and processing speed.
And from a completely different segment — we worked on a project for a global e-commerce company, where we contributed to developing a machine-learning-based ranking engine. The algorithm adjusts product order based on relevance and business rules across different countries, improving search metrics and increasing customer conversion rates.
How do you approach such projects — a large project from the start, or a smaller prototype first?
We’ve found that starting with a PoC works best. At the beginning, we set clear goals, milestones, and KPIs with the client — whether we’re targeting time savings, accuracy, SLA compliance, or shorter information retrieval time.
Within a fewweeks we create a functional prototype and let internal employees on the clientside test it (typically 5–6 testers) who work with it and evaluate thesolution. The first version might have around 60% success rate.
Then we go through their feedback and adjust the solution based on analysis and evaluation. After that we deploy another version and test again. This way we can get up to 97% accuracy — as in the PoC mentioned above.
It’s much easier to talk about AI when you can show the client: “Look, this cut your internal search time by two hours per day,” rather than staying at “AI is trendy.”
So your role is to connect the worlds of business and IT?
Yes. I’m the person who translates business requirements into the language of developers and architects. I identify the problem with the client, together we turn it into a concrete AI use case, and then I work with the team to design the solution, manage the PoC and the delivery. I also ensure the solution makes real business sense and that the client knows exactly what they're getting from AI.
Generative AI and Agents in Practice: Benefits vs. Client Concerns
There’s a lot of talk about generative AI and agents. How do you perceive them in practice?
I see generative AI today as a technology that is no longer experimental but a stable part of enterprise ecosystems. We primarily use it in platforms like Azure OpenAI or AWS Bedrock. You don’t have to program everything from scratch — you can work with configuration, system prompts, and personalization. Compared to the early days of GPT tools, today’s models are significantly more accurate, but the quality of answers always depends on how well the solution is designed, what data it’s based on, and what infrastructure it runs on.
Agents are, in my view, the next natural step. It’s no longer just about generating text, but about autonomously completing specific tasks that a human typically performs. A typical example is our background-check agent: the user simply enters a company ID and the agent assesses the company’s risk level, checks public sources, identifies key people, extracts financial indicators, and finally compiles the entire investigation into a structured PDF. What would take people days, the agent handles in minutes.
Looking ahead, I see agent-based approaches becoming standard anywhere repetitive, process-driven, or data-heavy work exists — for example scheduling meetings, transcribing data, or repeating routine tasks. And not only for clients. I can imagine an internal agent who, based on a simple command, organizes a meeting for me, finds a free meeting room, prepares documents, and later processes the outputs. I believe agents will be what makes AI a normal part of everyday work.
Do clients have real concerns about AI or agents?
Yes ,questions about concerns typically revolve around three main areas. First is data security. Clients want to be sure their internal information stays within their environment and doesn’t “leak” anywhere. We address this through dedicated tenants, private networks, encryption, and security testing. Our AI tools run either in the client’s infrastructure or in our secured environment.
Another areaclients ask about is hallucinations and answer consistency. I often encounterthe expectation that AI must answer identically to the same question everytime. But that’s not how generative models work — answers may be phraseddifferently while still being correct. We mitigate hallucinations by trainingon specific data and intensive testing — we continuously “show” the modelwhat’s right and what’s wrong.
Sometimesclients worry that AI will take their jobs. Especially people who work with thesolution — backoffice staff, operators, testers. They feel like they’re testinga tool that’s meant to replace them. We explain that the goal isn’t reducingheadcount, but easing the burden of repetitive manual work that consumesenormous amounts of time. That’s also why we end projects with training andeducation, so it’s clear how the solution helps them specifically.
AI Not Just for Clients: Tools That Help in Everyday Life
How do you personally use AI?
AI is integrated into my everyday functioning. At work I use it as an analytical partner that helps me structure problems, quickly get oriented, compare scenarios, or validate the logic of documents I’m preparing. I also use it a lot for operational tasks — AI drafts presentations, summarizes long documents, compares versions, and evaluates client feedback. And I also use AI for planning: it can optimize my calendar or define priorities for the coming week.
In my personal life, I use AI very practically. It helps with trip planning, finances, decision-making, or preparing training plans. For sports, it works with data from Garmin or Strava and suggests workouts based on current performance.
I also create my own agents — for example for studying. For my final exams, I created an agent into which I uploaded materials (even 400 pages), and it tested me. It told me where I was answering too long, where my gaps were, what I couldn’t explain, what I should focus on, and so on.
The 5 Most Useful AI Tools According to Anička Plašek
Chat GPT (“chat”) / custom GPT agents
– structuring problems, brainstorming, preparing materials, text summaries, vacation itineraries, training plans, creating shopping lists
Perplexity AI
– factual search from verified sources, research and current information, comparison of opinions, trend summaries
Notion AI
– work organization, task management, project notes, automatic meeting summaries
DALL-E 3
– creating visuals, illustrations, concepts, redesigns, social media posts, presentations, creative assets
Claude 3 (Anthropic)
– work with long texts, detailed analyses, understanding documents, an alternative to Chat GPT for complex content and high-precision formulations
How did you actually get to Trask?
I graduated from the University of Economics in Prague, FIS, and during my studies I had internships at Deloitte and Deutsche Börse Group. Deloitte was a great learning experience, but also a very process-heavy environment — either there was almost no work or, on the contrary, twelve-hour days with the prospect of career advancement maybe in ten years. I’m someone who can work very intensively and go beyond expectations, but I also need to see meaning, pace, and impact — and that was missing there.
I got to Trask completely by accident. I told this whole experience to Martin Rýznar, my classmate from university who works at Trask. He said, “Come join us.” The nextday I sent him my CV, it landed with Lucka Slavíčková, the then Senior Manager of Data Science, followed by an interview and very quickly an offer.
I like to say I got into it “like the blind hen finding a grain,” but looking back, I see it as one of the best professional decisions I’ve ever made.
Anna Plašek
Anička has been at Trask for more than three years, specializing in project management in the field of artificial intelligence and Data Science. Before joining Trask, she gained experience in Deloitte’s Tax & Legal CIT internship program and in the analytical team at Deutsche Börse Group.
She graduated from the University of Economics in Prague at the Faculty of Informatics and Statistics (FIS VŠE). She focuses on topics such as AI, IS/ICT, data mining, data analysis, project management, and business intelligence. In her work, she combines technological understanding with practical business impact and actively contributes to the development of AI competencies within the DS department.
