🎤 Data Sanity Talks: Final Call for AI Projects and Discussions!

🚀 AI Projects: Showcase your AI pet project, research, or initiative during the poster session! Even if it’s small or in the early stages of development, it’s a great opportunity to gather feedback and find collaborators.

🎙 AI Discussions: Lead a discussion on your favorite AI topic — whether technical or high-level! Help exchange ideas and discover new perspectives on an important (or just fun) ML topic.

✨ Apply & Learn More: https://datasanity.dev/
📅 Deadline: May 8

Need more info? Reach out to @yarrrro.

⭐️ Join this event on LinkedIn
⭐️ Follow us on Instagram

We're excited to support AI Week in Belgrade this year as a community partner!

AI Week is all about spreading knowledge and connecting curious minds — perfectly in line with our mission at Data Sanity.

That's why we're kicking things off with Data Sanity Talks on Monday and Tuesday (June 2 & 3). Stay tuned: public registration opens on May 11! 🚀

Check out all AI Week events: https://startit.rs/ai-week/

🎤 Data Sanity Talks: Public Registration Is Open!

🗓 When: June 2 + June 3, 18:00 — 22:00
📍 Where: Belgrade, Startit Centar

Are you ready for two evenings of AI talks, projects, and discussions? Join our free event to gain industry insights, connect with professionals and executives, and enjoy free snacks! 🥐

Visit our website to see the current lineup of speakers, and follow us on social media for more details on the event activities.

🚀 Register to secure your spot
⭐️ Learn More: https://datasanity.dev/
⭐️ Join this event on LinkedIn
⭐️ Follow us on Instagram

Want to support us? Reach out to @yarrrro.

🎙 Building Better Datasets: Image & Video Quality at Scale

Speaker: Ivan Drokin, Applied ML Lead @ Moonvalley.ai

High-quality data is the backbone of effective generative models — especially for text2image and text2video systems. In this talk, we’ll explore state-of-the-art methods for evaluating and filtering image and video datasets, including large-scale deduplication techniques that work across billions of samples. Ivan will also share hands-on lessons from working with public datasets and highlight common pitfalls and best practices for dataset curation at scale.

🗓 When: June 2 + June 3, 18:00 — 22:00
📍 Where: Belgrade, Startit Centar
🚀 Register & Learn more: https://datasanity.dev/

🎙 Creative Assistants: Embedding AI in Design & Art

Speaker: Veljko Ćirić, Creative Technologist & Architect

How can AI become a true creative partner rather than just a tool for automation? In this talk, Veljko Ćirić explores how analog AI inference, custom LoRA training, and artistic thinking can transform AI into an adaptive, intuitive assistant. Drawing on his background in architecture and design — as well as research in Human-AI interaction — he presents a vision of AI that senses, responds, and co-creates alongside humans.

🗓 When: June 2 + June 3, 18:00 — 22:00
📍 Where: Belgrade, Startit Centar
🚀 Register & Learn more: https://datasanity.dev/

🎙 Post-training LLMs: Smarter Algorithms & Rewards

Speaker: Kirill Tyshchuk, AI Engineer @ Perplexity AI

What’s next for fine-tuning large language models? This talk dives into the fast-moving world of LLM alignment, comparing classic techniques like PPO and DPO with cutting-edge alternatives such as RLOO, CGPO, GRPO, and DAPO. Kirill Tyshchuk will unpack how verifiable reward signals are driving breakthroughs in math reasoning, code generation, and structured output quality. Expect a practical, insight-driven tour through real-world challenges and industry applications.

🗓 When: June 2 + June 3, 18:00 — 22:00
📍 Where: Belgrade, Startit Centar
🚀 Register & Learn more: https://datasanity.dev/

🎙 Agentic Product Design: What Works, What Fails, and Why

Speaker: Arsenii Shkunkov, AI/ML Engineer @ Raw Ventures

AI products are evolving from static tools into dynamic systems with autonomous behavior — but what truly makes an AI system _agentic_? In this talk, Arsenii explores what sets successful AI agents apart from those that underperform or fail. He’ll break down common design pitfalls, from poorly defined scopes to misaligned autonomy, and share practical strategies for defining the right boundaries and balancing independence with human oversight. If you’re building — or planning to build — AI agents, this talk will offer the insights you need to do it right.

🗓 When: June 2 + June 3, 18:00 — 22:00
📍 Where: Belgrade, Startit Centar
🚀 Register & Learn more: https://datasanity.dev/

🎙 Launching AI Startups: From First Prototype to First Clients

Speaker: Igor Dostanić, Co-Founder and CTO @ Luccid.ai

What does it take to go from idea to first paying clients in the AI space? Igor Dostanić shares real-world lessons on building MVPs that actually work, finding product-market fit, and talking to clients in traditional industries like construction. Learn how to build lean, earn trust, navigate investor conversations, and handle messy real-world data — all while avoiding the trap of building tech for tech’s sake.

🗓 When: June 2 + June 3, 18:00 — 22:00
📍 Where: Belgrade, Startit Centar
🚀 Register & Learn more: https://datasanity.dev/

🎙 Interactive Latent Diffusion: Steering Text-to-Image Models

Speaker: Nick Knizev, Co-Founder @ Wizium.ai, Ex-Meta

What if you could collaborate with an AI to generate images — not just prompt it and hope for the best? In this talk, Nick Knizev shares his Best Paper–nominated work on Interactive Latent Diffusion Models (IELDM), a new approach that lets users steer text-to-image generation in real time. This allows users to guide image generation by selecting preferred outputs and interacting with specific image regions, helping the model learn and adapt to user intent. These targeted refinements and image recombination help to reduce the trial-and-error, showing strong performance even with complex prompts that typically challenge diffusion models.

🗓 When: June 2 + June 3, 18:00 — 22:00
📍 Where: Belgrade, Startit Centar
🚀 Register & Learn more: https://datasanity.dev/

🎙 Managing Data Scientists: A Story of Uncertainty, AGI & Sand Frogs

Speaker: Maria Matvejeva, DS Manager @ AdTech Holding

Managing a data science team isn’t like managing developers. It’s a constant balancing act — between innovation and delivery, research and results, metrics and business value. In this talk, Maria shares lessons from her transition into data science leadership: why A/B tests can mislead, why great models don’t always drive impact, and how to keep a team motivated without drifting into research rabbit holes. Expect real stories, practical frameworks, and a few sand frogs along the way.

🗓 When: June 2 + June 3, 18:00 — 22:00
📍 Where: Belgrade, Startit Centar
🚀 Register & Learn more: https://datasanity.dev/

🎙 How We Use AI Tools: Mental Models & Misconceptions

Speaker: Ilya Zakharov, Senior Researcher @ JetBrains

How do our assumptions shape the way we use AI tools — and misuse them? In this talk, Ilya Zakharov explores the mental models that guide how experts and non-experts alike interact with AI systems. Drawing from cognitive science and real-world examples, he contrasts expert mental models with those of the broader public, uncovering key gaps that impact usability, trust, and outcomes. Learn how better mental models can lead to more effective — and safer — AI use.

🗓 When: June 2 + June 3, 18:00 — 22:00
📍 Where: Belgrade, Startit Centar
🚀 Register & Learn more: https://datasanity.dev/

🎙 Talk: Hacking the Prompt: AI Security in the Era of LLMs

Speaker: Miloš Božić, CEO & Founder @ Pyxero.ai

As Large Language Models (LLMs) power more tools and platforms, the risks of misuse and manipulation are growing fast. In this talk, Miloš Božić shares insights from extensive testing of open-source LLMs — exposing how techniques like prompt injection, system impersonation, and chain-of-thought hijacking can bypass safety filters and exploit model behavior. You’ll learn where current defenses fall short, and what needs to change to make AI more secure and trustworthy.

🗓 When: June 2 + June 3, 18:00 — 22:00
📍 Where: Belgrade, Startit Centar
🚀 Register & Learn More: https://datasanity.dev