CO-LOCATED EVENTS

Hotel Metropol, Belgrade

18th-22nd November 2024

DSC SCHEDULE

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.

Ivo Andric A

NLP

Registration

09:30

 –

10:00

Registration

Overcoming Loneliness with LLM Dating Assistant

NLP
Business talk
Intermediate
In today’s AI-driven world, every industry is racing to harness the power of AI and LLMs, and social discovery is no different. Social Discovery Group builds innovative dating and entertainment apps aiming to solve the problem of loneliness. Have you ever felt hesitant about what to say to a match? We’re building an AI assistant that helps our users communicate more effectively with potential matches, overcoming shyness and fear of rejection. This talk will describe our approach to developing, testing, and iterating on this LLM-powered assistant – from data analysis and model training to A/B testing and real-world deployment.
10:00

 –

10:30

Overcoming Loneliness with LLM Dating Assistant

Dani El-Ayyass
Head of AI Transformation @ Social Discovery Group

Where there’s a will, there’s a way: The mad genius of LLMs as classifiers

NLP
Technical talk
Advanced
Have you ever wanted to build a trillion-parameter labelling machine? And by that I mean, would you use a Large Language Model to solve a classification problem?

It’s an unlikely fit for an LLM. Classifiers typically need to be fast, accurate, and interpretable. LLMs are slow, random black-boxes. Classifiers need to output a single label. LLMs never stop talking.

And yet, there are good reasons to use LLMs for such tasks, and emerging architectures and techniques to do so. Many real-world use cases need a classifier—whether it’s their end goal or just a step in a larger process—and many data and product development teams will soon find themselves wondering, could GPT handle that?

If that sounds like you, then join me for a deep dive into how we’re tackling this issue while building a next generation conversational assistant at Switzerland’s largest telecommunications provider. I’ll cover:
– Why reducing language to labels is conceptually problematic, but practically useful
– Why classification with LLMs is a weird idea, but you might want to do it anyway
– Possible approaches and architectures
– Customer intent detection as a real-world use case
– Universal challenges, and our lessons learned so far
10:30

 –

11:00

Where there’s a will, there’s a way: The mad genius of LLMs as classifiers

Katherine Munro
Data Scientist, Conversational AI Engineer @ Swisscom

Leveraging multimodal language model for AI native wireless systems

NLP
11:00

 –

11:30

Leveraging multimodal language model for AI native wireless systems

Ali Mokh
Researcher Concept AI @ Ericsson

Coffee Break

11:30

 –

12:00

Coffee Break

Keynote: Fighting AI with AI

Keynote
Technical talk, Business talk, Research talk, Educational talk
Intermediate
How do we tell what’s true and what’s not? In a world where artificial intelligence presents both promises and perils, we find ourselves in a unique position: harnessing the very technology that threatens to undermine truth and authenticity to defend against it. In this presentation, we’ll discuss How Do Deepfakes Work? Real-life examples of fraud perpetrated using deepfake techniques. And what can be done? [Deep Fake Voice Detection (Behavioral Signals)]
12:00

 –

12:45

Keynote: Fighting AI with AI

Rana Gujral
CEO @ Behavioral Signals

Keynote: Data-Driven Success: Securing Executive Buy-In for a Game-Changing Data Strategy

Keynote
Business talk
Beginner to Intermediate
In today’s data-driven landscape, aligning your data strategy with organizational objectives is critical for achieving sustainable growth and gaining competitive advantage. This session explores the process of developing a robust data strategy that not only supports and enhances your company’s strategic goals but also delivers tangible value across the enterprise. Learn how to effectively communicate the importance of data initiatives to senior leadership, ensuring executive buy-in and support. We will delve into best practices for identifying key performance indicators, integrating data insights into decision-making processes, and demonstrating ROI to stakeholders. Join us to discover actionable steps to create a cohesive data strategy that drives business success and secures the commitment of your executive team.
12:45

 –

13:30

Keynote: Data-Driven Success: Securing Executive Buy-In for a Game-Changing Data Strategy

Amie Bright
VP Enterprise Data & Insights @ GitLab

Lunch Break

13:30

 –

14:30

Lunch Break

Keynote

Keynote
14:30

 –

15:15

Keynote

Karen de Sousa Pesse
Senior Executive @ Salesforce

How AI Automates International Trade, Boosts Profitability – and Where It Fails: A Practical Talk

NLP
Technical talk, Business talk
Intermediate
In his insightful presentation, Peter Piehler will explore how AI supports critical processes in international trade, helping companies to future-proof their operations, minimize risks, enable transactions, and maximize profits. He will also shed light on the current limitations of AI and the reasons behind its failures. The talk will delve into why AI does not replace natural language processing (NLP) and why NLP is essential for delivering high-quality performance. Mr. Piehler will discuss the implementation of quality assurance and plausibility checks for AI-supported responses. The advantages of argumentative AI over traditional prompt-and-respond approaches will be examined. Technologies will be introduced, and experiences shared, with the hope of passing on the enthusiasm for open-source AI. To conclude, Mr. Piehler will venture a forecast on how technology is transforming the software industry market, specifically for software in international trade.
15:15

 –

15:45

How AI Automates International Trade, Boosts Profitability – and Where It Fails: A Practical Talk

Peter Piehler
CEO @ PST.AG

Coffee Break

15:45

 –

16:00

Coffee Break

Empowering Sales with Intelligent BI Agents: DWH, LLM, and RAG Integration

NLP
Technical talk, Business talk
Intermediate
In today’s data-driven business landscape, traditional BI tools often fall short, requiring specialized knowledge and struggling with complex queries on large-scale databases. Our innovative approach bridges this gap by seamlessly integrating Data Warehouses, Large Language Models (LLMs), and Retrieval-Augmented Generation (RAG).

This cutting-edge system empowers sales professionals to interact with complex data using natural language, eliminating the need for SQL expertise. By leveraging AI, we’re enabling rapid, context-aware insights, allowing users to access and interpret data more efficiently.

Join us to explore how this integration is transforming the way sales teams make data-driven decisions, marking a leap forward in accessible, intelligent business analytics.
16:00

 –

16:30

Empowering Sales with Intelligent BI Agents: DWH, LLM, and RAG Integration

Simun Sunjic & Lovro Matosevic
Senior Software Developer @ Atomic Intelligence & ML Engineer @ Atomic Intelligence

VerifAI: Biomedical Generative Question-Answering engine with verifiable answers

NLP
Technical talk, Research talk
Intermediate
In this talk, we will present our work on VerifAI system, an open-source biomedical question-answering system with unique mechanism to verify answers and detect hallucinations in generated answers. Namely, sciences, especially life sciences, have a low tolerance for non-factual information, and therefore many practitioners have been skeptical of using available tools, such as ChatGPT. While providing references to the information sources is a step in the right direction, it may not be enough, and even referenced generated answer may contain hallucinations. Therefore, we have developed a set of methods that on top of advanced RAG, combining lexical and semantic search capabilities and fine-tuning LLM with performance efficient fine-tuning method, such as LoRA, verify the answer and detect any remaining hallucinations. In order to increase efficiency, we have used quantization to improve latency and decrease hardware requirements for hosting the product. We will in detail discuss these techniques. The code, models and datasets generated during the project have been published in open-source and open science manner.
16:30

 –

17:00

VerifAI: Biomedical Generative Question-Answering engine with verifiable answers

Nikola Milosevic
Science Fellow @ Bayer

Official Closing

17:00

 –

17:15

Official Closing

Ivo Andric B

Applied Data Science,

Data Platforms

Registration

09:30

 –

10:00

Registration

How to Encourage Children to Read Books? LLMs for Education and Fun!

Applied Data Science
Technical talk
Beginner to Intermediate
Children who read books regularly grow up to be better-educated adults with higher earning potential and stronger emotional intelligence. But how can we encourage them to read books? In this talk, you will learn how SoftwareOne leveraged Large Language Models and AWS Cloud to create the Book Aligned Activities AI Engine for Worldreader, a US-based non-profit organization. This AI Engine allows Worldreader to create engaging and creative book activities for children in Kiswahili and English, reducing the development time from weeks to minutes.
10:00

 –

10:30

How to Encourage Children to Read Books? LLMs for Education and Fun!

Ewelina Kucal & Maciej Dziezyc
Data Scientist @ SoftwareOne & Senior Data Scientist @ SoftwareOne

Predicting Retention on the BBC Sounds Service Using Historic User Dynamics

Applied Data Science
Technical talk, Business talk
Intermediate
Accurately predicting customers’ propensities to remain with a service allows businesses to stage informed and targeted marketing campaigns in the hope of retaining individuals for longer and to reduce subscriber attrition. Moreover, retention probability can be used in downstream tasks as a metric in A/B tests (e.g. in evaluating recommender systems). Launched 6 years ago, BBC Sounds is a streaming and download service which provides users with live radio, on demand audio and podcasts. With its weekly audience figure peaking at over 5 million users in the first quarter of 2024, the BBC would greatly benefit from having a way of identifying its most loyal listeners who are likely to stay with the Sounds service at different distances into the future. Often times, retention modellers focus on engineering and employing static features (e.g. age range, income, location, etc.) to estimate the probability of someone staying with or leaving a service. In this work, we model a user churning from or remaining with BBC Sounds as a simple binary classification problem, but use a flexible framework to leverage the information included in a user’s historical temporal features. For example, we use the number of minutes consumed, as a timeseries, leading up to a retention/churn event, amongst other dynamic attributes to predict the probability of an account continuing to use the service. We investigate the efficacy of using such temporal numerical features alongside and without static features to predict the probability of retaining a user at different intervals into the future (N=7, 30 and 90 days). We evaluate the performance of different temporal machine learning architectures (convolutional and/or recurrent neural networks) and compare them with static alternatives (e.g. XGBoost). We also explore how different feature engineering techniques help or hinder model performance on held out (test) data. Our results elucidate the optimal way of predicting retention for BBC Sounds, alongside the optimal amount of temporal history required to maximise model performance.
10:30

 –

11:00

Predicting Retention on the BBC Sounds Service Using Historic User Dynamics

Ryan Timms
Senior Data Scientist @ BBC

Building up the Bosch Semantic Data Lake

Applied Data Science
Technical talk
Advanced
Creating the Bosch Semantic Data Lake as a solid foundation that synergizes Data Lakes and Knowledge Layers as Semantic Data Lake, that offers scalable data management platform for the flexible analysis (e.g., reporting, machine learning, stream analytics) of all kinds of data (e.g., batch & stream, structured & unstructured, raw & aggregated) with Semantic Metadata Management and Semantic Data Modelling, making it a Semantic Data Lake suited for large-scale batch/stream storage and processing that is scalable and cost efficient.
11:00

 –

11:30

Building up the Bosch Semantic Data Lake

Sofija Pervulov
Technology Data Steward, Data Architect @ Bosch Digital

Coffee Break

11:30

 –

12:00

Coffee Break

Black

12:00

 –

12:45

Black

Black

12:45

 –

13:30

Black

Lunch Break

13:30

 –

14:30

Lunch Break

Panel

Panel
14:30

 –

15:15

Panel

“Moderator: Marija Novakovic, Panelist: Jovana Petrovic
Program Director @ DSC Franchise & Game Math Designer @ Playstudios Europe

The path to Effective Data Migration – Overcoming Challenges and Ensuring Success

Data Platforms
Business talk, Technical talk
Beginner to Intermediate
This talk will provide actionable insights on managing data migration projects. We will be discussing lessons learned from multiple migration to public or hybrid cloud. We will not be focusing on implementation details. Instead we will cover key strategies for ensuring a smooth migration, addressing common challenges and optimizing costs. Key Points: Accelerating data analysts’ transition to self-service infrastructure. Upskilling and training initiatives for data scientists. Implementing audit logs, Ensuring GDPR compliance and security. Why it’s important and why it’s tricky. Enhancing data science workflows with MLOps tools and processes. Tool selection based on team skillsets. Various migration types aka why lift and shift is not optimal. Vendor lock-in. This session is designed for CTOs, CDOs, and data leaders looking to streamline their migration processes.
15:15

 –

15:45

The path to Effective Data Migration – Overcoming Challenges and Ensuring Success

Marcin Szymaniuk
CEO @ TantusData

Coffee Break

15:45

 –

16:00

Coffee Break

Our journey to harden the security and compliance of the Data Platform

Data Platforms
Technical talk
Intermediate to Advanced
Talk about our journey to harden the security and compliance of the Data Platform with the steps and learnings we have to date.
16:00

 –

16:30

Our journey to harden the security and compliance of the Data Platform

Dennis van Rooijen
Director Data Platform @ GitLab

Personalized Insights and Engagements using Customer Data Platforms for Enterprises to make data driven decisions

Data Platforms
Technical talk, Business talk
Intermediate to Advanced
How Personalized Insights and Engagements using Customer Data Platforms and Enterprise Data Hubs are of utmost importance for Enterprises to make data driven decisions to be a market leader
16:30

 –

17:00

Personalized Insights and Engagements using Customer Data Platforms for Enterprises to make data driven decisions

Pratul Kumar Chakravarty
Senior Practice Manager @ Adobe Systems

LavandER

Responsible AI,

Transforming industries with AI

Registration

09:30

 –

10:00

Registration

Open Data for Everybody: Social Action, Peace Tech, and Making Open Data Outreach More ‘Open’ for Communities

Responsible AI
Educational talk
Beginner to Intermediate
What if I told you about something that could empower our third sector and activists to enhance their capacity? From gathering evidence for funding tenders to campaigning for crucial social issues and much more? It’s called open data, yet many in social action remain unaware of it. Primarily shaped by corporate entities, open data seems tailored only for technologists, alienating the third sector. But in reality, it’s a powerful tool for social change, bolstering civil society, and creating resilient communities.
In this presentation, Nathan Coyle delves into the findings from his book, highlighting the obstacles that have prevented the widespread use of open data. His research uncovers the systemic issues within governmental institutions that lead to ineffective policies. Nathan will also cover how we can bolster peacebuilding initiatives, focusing on their specific requirements and demonstrating how we can shape data governance to enhance the accessibility of open data.
We need to shift our perspective on open data, viewing it as a powerful tool for social good that can drive positive change within our communities. By doing so, we can empower activists and the third sector, helping them build capacity, secure necessary funding, and influence policy-making processes.
The session will also look at peace technology, examining its applications and the stories that need to be told. Furthermore, we will explore how the social tech community in Serbia can advocate for more open and accessible data from providers, thereby supporting social activists and peacebuilding organisations.
Nathan’s first book, Open Data for Everybody: Using Open Data for Social Good, is now available through Routledge: https://www.routledge.com/Open-Data-for-Everybody-Using-Open-Data-for-Social-Good/Coyle/p/book/9781032715049.
10:00

 –

10:30

Open Data for Everybody: Social Action, Peace Tech, and Making Open Data Outreach More ‘Open’ for Communities

Nathan Coyle
Senior Peace Tech Advisor @ Austrian Centre for Peace

2025: year of Ai dilemma – ethics, regulations and innovations

Responsible AI
Educational talk
Intermediate
Over the past decade, machine learning has seamlessly integrated into every facet of our daily lives, revolutionizing how tasks are executed with unprecedented efficiency and cost-effectiveness. This surge in machine autonomy, however, has brought forth a conundrum reminiscent of the Oppenheimer Dilemma – the profound responsibility accompanying great technological power. The maxim “with great power comes great responsibility” resonates ever more deeply in this era of accessible machine learning, where even novice developers can wield immense influence. From granting access to transformative financial resources to shaping social media narratives that can polarize beliefs, the scope of machine influence is vast and impactful. As AI technologies burgeon and come under increased scrutiny, the landscape of AI development undergoes a profound shift. The entire data science lifecycle must now integrate principles of responsible AI, encompassing fairness, accountability, privacy, and transparency. Professionals adept at implementing these principles are increasingly sought after by employers. Furthermore, it’s imperative to address issues like hallucination, monetization, robustness, reliability, and attribution within a broader Responsible AI framework. Moreover, principles like human-centricity, planetary inclusivity, safety, and societal well-being must not be overlooked. This imperative extends beyond current technologies to include emerging frontiers like General AI. Principles of fairness, explainability, transparency, robustness, stability, safety, and privacy must be rigorously upheld in the development of General AI. Embedding these foundational pillars ensures that the potential of these advanced systems is harnessed ethically for the collective good of humanity. Prioritizing these principles not only mitigates potential risks but also cultivates an environment where Genl AI can flourish in alignment with our values and aspirations for a better future. Beside the above, the main dilemma that will be at the centrestage will revolve around Regulations vs. Innovation, Global Coordination, Wealth Inequality, Reliability, Social Impact and Control. These dilemmas are interconnected and multifaceted, requiring collaborative efforts from technologists, policymakers, ethicists, and the public to navigate. The year 2025 is seen as a critical juncture as the pace of AI development accelerates and its integration into various aspects of society becomes more pronounced.
10:30

 –

11:00

2025: year of Ai dilemma – ethics, regulations and innovations

Sray Agarwal
Director Responsible AI @ Fractal

TBA

Responsible AI
11:00

 –

11:30

TBA

Arnault Ioualalen
CEO and R&D director @ Numalis

Coffee Break

11:30

 –

12:00

Coffee Break

Black

12:00

 –

12:45

Black

Black

12:45

 –

13:30

Black

Lunch Break

13:30

 –

14:30

Lunch Break

Panel

Panel
14:30

 –

15:15

Panel

Application of AI technologies in nuclear power plants

Transforming industries with AI
Technical talk
Intermediate to Advanced
Example of several AI use-cases deployed and used in Slovak nuclear power plants improving maintenance and operation. Topics would include experience of deployment, human interaction & rugulator aspect for AI solutions.
15:15

 –

15:45

Application of AI technologies in nuclear power plants

Peter Kertys & Maros Buban
Head of Data Science @ Slovenské elektrárne & Solution Architect @ Slovenské elektrárne

Coffee Break

15:45

 –

16:00

Coffee Break

TIS Group’s AI Journey from Healthcare to Agriculture and then back to the roots

Transforming industries with AI
Business Talk
Beginner to Intermediate
• Why does TIS.AI interact with infants? (sendd.eu)
• How do vineyards interact with TIS.AI? (vineyardangel.com)
• What is TIS.AI official LLM language?
• Where are our business roots?
16:00

 –

16:30

TIS Group’s AI Journey from Healthcare to Agriculture and then back to the roots

Tomislav Strgar
Head of AI Department @ TIS Group

TBA

Transforming industries with AI
16:30

 –

17:00

TBA

Orsalia Andreou
Data Scientist @ Erste Group

Tesla B

AI & Data Product Development

Registration

09:30

 –

10:00

Registration

TBA

AI & Data Product Development
10:00

 –

10:30

TBA

Luka Anicin
CEO & Artificial Intelligence Consultant @ Datablooz

Future of the future: creating an AI product beyond hype

AI & Data Product Development
Technical talk, Business talk
Intermediate
When a company creates one of a bunch of AI products, what should distinguish it from competitors? We know the initial AI hype is behind us, the real things are happening now and now everyone is asking: what`s next?

Let’s talk about GitLab Duo, our AI baby. We ramped it up in an aggressively narrow timeframe, with a vision to provide the highest quality experience for our users. At the moment, many questions are opened about the future of its development.

Join me to examine the landscape of AI products in the DevSecOps world and the trajectory of future expansion. We are delighted to share the experience we have on the top of the product with more than 30 million users.
1. Why do we jump into the unknown area with LLM and AI development?
2. What is the approach that will become a success story for AI product development – yes you guess, it is an iterative approach
3. How to leverage dogfooding – and what is dogfooding – Tip: it is not about the dog or the food
4. When to pivot your development without regret – yes, it is because of FinOps, but not only that
5. How to measure your success – people often forget to measure what they are doing and are they will reach the goal

I promise – no fluff, no buzzwords. Solely a first-face story from the GitLab Data Team member, created from sweet and pain from the world leader of DevSecOps platform creator, remote work champion and one of the most transparent company in history.
10:30

 –

11:00

Future of the future: creating an AI product beyond hype

Radovan Bacovic
Staff Data Engineer @ GitLab

How to effectively manage AI/ML projects

AI & Data Product Development
Use case
Advanced
Understanding the complexities of AI project management is crucial as it directly impacts the alignment of AI initiatives with organizational goals. In this session, Ana Stojković Knežević will explore the specialized frameworks necessary for managing AI projects effectively. Participants will learn about Data Driven Scrum (DDS), a methodology tailored for AI that enhances project efficiency through flexibility and iterative learning. Key takeaways include strategies for balancing technical and business requirements, the importance of staying updated with AI trends, and practical steps for implementing DDS.
11:00

 –

11:30

How to effectively manage AI/ML projects

Ana Stojkovic Knezevic
Project Manager @ Protech

Coffee Break

11:30

 –

12:00

Coffee Break

Keynote: The Art of Telling Stories with Data

Keynote
Join us for a dynamic and entertaining keynote session on Data Storytelling with Gulrez Khan, Data Science leader at PayPal and Award Winning Author. Gulrez is known for infusing his presentations with humor and personal stories, making the learning experience both engaging and enjoyable. You’ll be inspired by Gulrez’s insights and experience as he guides you through the process of turning numbers into narratives. Discover how to craft compelling stories that bring your data to life, and learn how to share your insights in a way that engages, educates, and inspires your audience.
12:00

 –

12:45

Keynote: The Art of Telling Stories with Data

Gulrez Khan
Data Science Leader @ PayPal & Award Winning Author

Keynote

Keynote
12:45

 –

13:30

Keynote

Lunch Break

13:30

 –

14:30

Lunch Break

Panel

14:30

 –

15:15

Panel

How Alice, our intelligent personal assistant, learned to understand people with speech impairments

AI & Data Product Development
Alice is Yandex’s virtual assistant, designed to answer questions, play music, assist with daily tasks, and manage smart home devices. Powered by the advanced YandexGPT neural network, Alice received a significant upgrade in July 2024, enhancing its ability to recognize voice requests from individuals with speech disorders, such as stuttering, cerebral palsy, stroke, or trauma.
This improvement was achieved through additional training of the neural network. During this process, Alice was exposed to hundreds of thousands of audio recordings provided by people with speech impairments. As a result, the accuracy gap between recognizing standard speech and speech with impairments was reduced by an average of 20%.
15:15

 –

15:45

How Alice, our intelligent personal assistant, learned to understand people with speech impairments

Anastasia Shapedko
Project lead, Technologies for Inclusion @ Yandex

Coffee Break

15:45

 –

16:00

Coffee Break

Balancing Personalization and Experimentation: The Art and Science of A/B Testing in a Customized World

AI & Data Product Development
As product teams strive to deliver personalized experiences that cater to individual user preferences, the challenge of maintaining standardized A/B testing conditions becomes more complex. This talk explores the intersection of personalization and A/B testing, presenting actionable frameworks and best practices for running experiments without compromising user experience or test validity.

From segmenting user groups to using multi-armed bandit algorithms, we’ll dive into advanced techniques that enable organizations to optimize personalized experiences while ensuring accurate, data-driven insights.

You’ll learn how to scale A/B testing efforts across multiple teams and product lines, overcome challenges like data inconsistencies, and create a culture of experimentation that balances the need for both personalization and standardization.

Whether you are a Product Manager, Data Scientist, or Marketing Professional, this session will equip you with the tools to seamlessly integrate personalization with your experimentation strategy for maximum impact.
16:00

 –

16:30

Balancing Personalization and Experimentation: The Art and Science of A/B Testing in a Customized World

Joy Chatterjee
Lead Data Scientist @ DCMN

AI Integration in Legacy Industries: Navigating the Path to AI Transformation

AI & Data Product Development
Transformational talk
Intermediate to Advanced
Explore the transformative power of AI in traditional industries, examining successful integrations and the unique challenges faced. This talk highlights how companies in sectors such as manufacturing, healthcare, and finance are leveraging AI to innovate and improve efficiency, while addressing the cultural, technical, and regulatory hurdles of adopting new technologies in established environments.
16:30

 –

17:00

AI Integration in Legacy Industries: Navigating the Path to AI Transformation

Marija Matokanovic
Product Owner Lead @ Andritz Digital Factory

* this program is not final and is subject to change, full schedule will be available soon