Meetup Grape Up x Addepto - How to Create a Chatbot: Application and Data
Building a chatbot? That part's easy.
Building one that actually works, at scale, securely, and with domain-specific knowledge? That's the real challenge.
If you've ever asked yourself:
- How do I build a chatbot that's more than just an API call - one that actually works in production?
- What are the common traps when working with LLMs, and how can I avoid them?
- How do I prepare data so my bot truly understands context, not just keywords?
- Why do basic RAG setups break in real-world use, and how do you make one that scales?
This meetup is for you.
We'll take a deep dive into the two pillars of building a reliable chatbot:
- Application - how to design, deploy, and monitor a chatbot system
- Data - how to feed it high-quality, contextual knowledge and validate the results
TALK 1: Chatbot application - easy to start, hard to master
with Damian Petrecki, AI Tech Lead @ Grape Up
From transformers and embeddings to context limits and prompt hacking, Damian will guide you through:
- Core concepts like tokenization, context windows, embeddings, and transformer architecture
- Backend-frontend architecture for scalable chatbot apps
- Tools like LangChain, Guardrails, Moderation API, Langfuse
- Common failure points - and how to avoid them
- Testing, monitoring, and maintaining your chatbot in production
Including a real use case: building a chatbot for car mechanics - what worked, what didn't, and why.
TALK 2: The RAG Paradox - When One-Fits-All Fits Nothing Well
with Tomasz Adamczyk, Data Scientist @ Addepto
Your chatbot is only as good as the data behind it. Tomasz will show how his team addresses that in ContextClue, Addepto's modular platform for RAG-based systems.
He'll cover:
- Why plug-and-play RAGs are a great start but not a long-term solution
- How to design modular, scalable architectures (and avoid fragile monoliths)
- Best practices for data cleaning, chunking, and context enrichment
- Validating performance using tools like ContextCheck
- A real-world success story: reducing 14 days of manual knowledge graph building to just 30 minutes through automation
Join us and meet others passionate about building smart, resilient, and domain-aware AI systems.
Snacks, good vibes, and post-talk chats included!
Where: Grape Up Office, Kraków ul. Żółkiewskiego 17a
When: June 12, 2025 at 18:00
Free entrance. The meetup will be held in Polish.
About the organizers:
Grape Up delivers modern and innovative software solutions, leveraging cloud computing and AI to help clients succeed in the digital landscape.
Grape Up Facebook Page
Grape Up Website
Addepto is a consulting company that delivers state-of-the-art AI and data-driven solutions.
Addepto Facebook Page
Addepto Website
About the speakers:
Damian Petrecki - AI Tech Lead at Grape Up. Damian is a seasoned software engineer with over a decade of experience delivering complex systems across industries, from EU-grade network platforms to next-gen banking apps and real estate tech with legal integration. Since 2019, he has been working with Grape Up, initially as a hands-on developer, and later as a technical lead for the automotive industry, solving some of the toughest challenges around Android Automotive OS deployments, modern IoT platforms for fleet management, and large-scale vehicle data processing.Damian is also Grape Up's internal LLM trainer, bridging the gap between AI and real-world automotive systems. He holds a Ph.D. in ICT, with a focus on vehicle control algorithms, and is known for pushing boundaries, mentoring teams, and turning complex problems into smart, scalable solutions.
Tomasz Adamczyk - Data Scientist at Addepto. Tomasz is a data scientist with over five years of professional experience, specialising in transforming complex data into actionable insights. He began his career in the telecommunications industry, where he helped bridge the gap between customers and products using data and AI-driven solutions. In 2024, Tomasz joined Addepto, where he contributes to a wide range of projects involving both classical machine learning and large language models (LLMs). He holds a Master's degree in Artificial Intelligence and Cognitive Science, reflecting his deep interest in how information is processed, both by machines and the human brain. This dual passion drives his work in building intelligent systems that not only deliver results but also align with human values and behaviour.