Data Sanity Talks

🚀 New Speaker at Data Sanity Talks 🚀 "AI Meets Healthcare: The Good, the Bad, and the Ugly" This is a keynote topic by Ivan Drokin, Head of Applied ML @ Moonvalley.ai, joining us from Montenegro. Ivan will talk about how AI is used in healthcare — what really works, what doesn’t, and the tricky parts in between. 🗓 October 4 | 12:00 — 22:00 📍 Belgrade, Startit Centar 🎟 Info & Tickets: https://datasanity.dev/

Data Sanity Talks Belgrade: Call 4 Posters There are 1-2 spots available on a first-come, first-served basis to: 1. Present your AI project during a poster session. Apply here! 2. Give a mini-talk or lead a discussion on a hot AI topic. Apply here! 🗓 October 4 | 12:00 — 22:00 📍 Belgrade, Startit Centar 🎟 Info & Tickets: datasanity.dev

💡 Training LLMs is insanely expensive — one run of Llama 3 cost an estimated $52M and took ~70 days on 16K GPUs. Every small optimization saves millions. 🖥 Longer context windows are a hot research race — they promise simpler workflows and infrastructure, but real performance gains are still limited. Big Tech and startups are betting heavily here. ⚡️ Reasoning costs are the next frontier — methods like speculative decoding, smaller distilled models, and even new chips (3,000 tokens/sec vs ~300 on GPUs) aim to make “thinking” cheaper. This is what we learned from the talk “LLMs Are Expensive — Here's How Big Tech Makes Them Less So” by Max Shaposhnikov, Research Engineer @ Tessl, ex-Applied Scientist @ Amazon Research Find other useful insights in the full video: https://youtu.be/DLsm2WEglSc?si=OHf6ENOWYq5-3ZaY

🚀 Data Sanity Talks Belgrade — October 4 is approaching! 🎤 Hear from global experts on the latest in AI & Machine Learning 👥 Connect with professionals shaping the future of tech & business 🗯 Join discussions on today’s most important (and fun!) AI topics 📍 Meet at Startit Centar Belgrade ➡ Sign up now! 🌍 datasanity.dev

🚀 First speaker coming to Data Sanity Talks 🚀 "Are Multimodal Transformers Disruptive Technology in Radiology AI?" Meet Yaroslav Nikulin — Head of AI @ Therapixel, artificial Intelligence for efficient breast cancer screening! He will share how multimodal transformers may mark radiology’s leap from "AI 1.0" to "AI 2.0", as Therapixel’s research shows both challenges and transformative opportunities ahead. 📅 Belgrade, October 4 | 12:00-22:00 See all speakers and buy tickets here: https://datasanity.dev/

How can you comply with EU AI laws and avoid jail time? ☠️ ☕️Yes, it’s a new drop from Data Sanity Talks! “Navigating the Enforcement of the EU AI Act: Practical Insights” by Michael Borrelli, Director @ AI & Partners 📺 Watch it here: https://youtu.be/RvPEHQlNuAE?si=y-HO_GJvUJK-UMow

3 hours left to grab an early-bird ticket for Data Sanity Talks! Why come? 🎤 Meet global experts presenting tech and business talks about AI/ML 👥 Network with AI professionals from different communities and industries 🗯 Participate in discussions about important and fun industry topics 🕊️ Get the lower price while it’s still live and share with a teammate who should join: https://datasanity.dev/

❗️Last chance! Early Bird tickets for Data Sanity Talks Belgrade fly away on Wednesday, September 3 🕊 Here's what awaits you: 🎤 10 talks by leading AI speakers ☕️ Curated AI discussions and project presentations 👥 150+ attendees ⏰ 8+ hours of quality content and networking ---+++--- 📍 Location: Belgrade, Startit Centar 🕐 Time: October 4, 12:00 — 22:00 See more details and grab your early bird ticket: https://datasanity.dev/

Hi folks! A quick reminder: our free workshop “Beyond RAG: From Retrieval to Agentic Systems” by Vladimir Ageev, Lead Data Scientist @ Epam Systems, is tomorrow! Vladimir will show the key steps in evolving RAG into agentic systems with focus on tool use, branching, and planning with LangGraph. 📌 August 30th, Startit Centar 📓The event is free, but don’t forget to register: https://lu.ma/shucn6o9

LLMs are moving into customer-facing workflows. Security can’t be an afterthought. Here are the essentials from “Hacking the Prompt: AI Security in the Era of LLMs”: 🚨 Prompt injection is the top LLM risk (low-skill, black-box). 🕵️‍♂️ Jailbreaks are just prompt injection (role‑play, impersonation, hidden tasks, Unicode/encoding). 🧭 System prompts matter: add explicit safety policy instructions for measurable gains. 🔐 Defense-in-depth + continuous red‑teaming after every change. How are you testing your LLMs today?

New drop from Data Sanity Talks in London! “LLM Super‑Teams: Modern Ensembling Strategies That Beat Any Single Giant” from Polina Proskura, Applied Scientist @ Amazon 🧠 Watch and learn about: 🟢 The challenge of uncertainty in real‑world LLM applications 🟢 Ensembling strategies to boost robustness and reliability 🟢 Methods to quantify and reduce model uncertainty Available on YouTube: https://youtu.be/eWXjSnxwSkQ?si=CYD2RXZcbkTaSZb7

Managing data scientists = managing uncertainty. Here's how to do it successfully: 🎯 Shape demand & educate: close the analyst vs. AGI gap; define target, data, and success upfront. 🧪➡️🚀 Baseline first, timebox research: ship something simple; explore within bounds to avoid sand‑frog over‑digging 🐸. 💰🔍 ROI gate early: stress‑test value vs. cost at the start and at reviews. Stop when the economics don’t hold. (top takeaways from “Managing Data Scientists: A Story of Uncertainty, AGI & Sand Frogs” by Maria Matvejeva, DS Manager @ AdTech Holding) Which one resonates most with your team right now?

How to transform your ordinary RAG into badass AI Agents? 💪 That's the topic of our new workshop: “Beyond RAG: From Retrieval to Agentic Systems”! 💽 Vladimir Ageev, Lead Data Scientist @ Epam Systems, will demonstrate the key steps in evolving RAG into agentic systems. He will focus on tool use, branching, and planning with LangGraph. After Vladimir's talk, we will discuss multi-agent architectures and their role in advanced reasoning! 🗓 Saturday, August 30, 11:00 — 13:00 📌 Belgrade, Startit Centar 📓 The event is free, but don’t forget to register: https://lu.ma/shucn6o9

How to make your work with AI most effective: 4 tips From the talk “How We Use AI Tools: Mental Models & Misconceptions” by Ilya Zakharov, Senior Researcher @ JetBrains: 1️⃣ Keep your picture of what AI can and can’t do up to date — it changes fast. 2️⃣ Treat it like a bright junior: give context, ask for a format, and iterate. 3️⃣ It does most of the work well but can fail in odd ways — keep an eye on it and pick tasks wisely. 4️⃣ How you talk to it matters; a little framing and tailoring to your level improves results. Keep learning, build a prompt library, and revisit your assumptions monthly.

Shipping AI that actually works is hard. Learn how pros do it. Like the sunrise, our Early Bird tickets won’t be around all day ☀️ Catch them while they last: https://datasanity.dev/

🚀 New workshop recording! Post-training LLMs: Smarter Algorithms & Rewards with Kirill Tyshchuk, AI Engineer @ Perplexity AI Large language models don’t stop at pre-training. The real magic happens in post-training and alignment. In this talk, Kirill explains: ✨ Why classic methods like PPO and DPO aren’t always enough ✨ What’s new with RLOO, CGPO, GRPO, and DAPO ✨ How verifiable reward signals are unlocking progress in math, coding, and structured outputs Watch the full session here

Last Call for Speakers ✨ Belgrade, October 4! ⏰ Deadline: Friday, 23:59 We extended our deadline for you, chads! Apply for a 15-40 min talk on AI, Data Science, or Data Engineering: 🚀 Technical: ML architectures, tooling, research, system design 🚀 Strategic: Leadership, trends, ethics 🔗 Learn more & Apply 🎯 First come, first served — only a couple spots left!

Come tomorrow to our free discussion! 🗒 Topic: "To AI or Not to AI? Integration & Adoption Case Study" with Paul Katurov. Paul will discuss his firsthand experience implementing AI within an organization, covering everything from finding high-impact opportunities to managing the transformation of employee workflows. Following his presentation, we'll start an open conversation about AI integration in workplace and personal settings. 📍 Startit Center, Savska 5, Belgrade 🕐 11 AM - 1 PM, 16.08. The event is free, but don’t forget to register: https://lu.ma/89komgof

New Data Sanity Talks — the biggest one yet! ☀️ It will be a full-day event packed with AI talks, discussions, and networking! Join us to learn best practices from leading AI experts and connect with professionals from diverse countries and fields. What's in store? 🗣 12 top speakers from the Data Science, Machine Learning, and AI fields 👥 150+ attendees 🧮 8 hours of high-quality content and meaningful networking 📅 October 4th, Belgrade Registration is open! Save with early bird pricing — ticket prices will increase as the event date approaches. https://datasanity.dev/

🔥 New fireside chat this Saturday! 🎆 Join us for "To AI or Not to AI? Integration & Adoption Case Study" with Paul Katurov! We’ll dive into the topic of AI adoption and how it’s being integrated into different aspects of work and life. Paul will share his own experience of bringing AI into a company, from identifying the areas where it can make the biggest impact to guiding how it transforms people’s work habits Afterwards, we’ll open the floor for a discussion on AI adoption in both professional and personal contexts. Everyone’s welcome to join the conversation! 📍 Startit Center, Savska 5, Belgrade 🕐 11 AM - 1 PM, 16.08. The event is free, but don’t forget to register: https://lu.ma/89komgof