Alexander Guschin, Engineer, Evidently AI

The presentation dives into integrating GenAI into products, using an audio guide for art museums as an example. It outlines ML tasks: reverse image search, description generation, and Text2Audio. Steps include data collection and model selection (like comparing OpenAI API vs LLMs you can run locally). It weighs API usage for quick prototyping against in-house solutions for data-heavy tasks, and suggests next steps you can take after your prototype is live to make it solve your specific challenges better.

Nemanja Petrović, Senior Software Engineer, B2ME

Explore the integration of Sentence-BERT (SBERT) to boost search functionalities in languages lacking comprehensive natural language processing (NLP) models. This presentation uncovers the process of training SBERT models tailored for empowering search systems in new language landscapes, addressing challenges and highlighting the potential impact on diverse linguistic communities. Join to learn about SBERT’s role in advancing information retrieval and accessibility in underrepresented languages.

Daniel Savenkov, Sr Machine Learning Engineer, JetBrains

The talk is about the details and motivation behind Local AI Code Completion by JetBrains that was released on the 4th of April 2024. I’ll start by talking about the current state of the most popular AI Code Completion tools. We’ll discuss the pros and cons of running language models for code in the cloud versus on the user’s machine. Then we’ll go through the main components of AI Code Completion which is much more than just a language model. We will discuss our approach to offline and online evaluation of AI Code Completion, prompt construction, and smart suggestion filtering. Also, we’ll discuss the rapid progress in efficient language model inference that we’re witnessing nowadays and the perspectives that local AI has given this progress. Finally, I’ll share some insights about the size of the team that made Local AI Code Completion for JetBrains IDEs happen, the model sizes that are currently used in production, and the cost of the models’ training.

Vladimir Ageev, Lead Data Scientist, Epam Systems

The talk is planned to be an overview of methods for inference speedup with specific examples and business motivation. For example: how to speed up text encoding with BERT, so embeddings are generated at service run time, not just retrieved from DB. This will include both software approach (e.g. onnx) and hardware (e.g. specialised AWS machines).

Andrija Sagić, Head of Digital Development, “Milutin Bojic” Library

Following the great and fast development of AI models we can note a weak support for low resource languages. For Serbian we have now YUGO GPT by Aleka Gordić, ORAO by JERTEH and fine tuning Whisper which are good examples of local efforts. But in other, so-called mainstream models (ChatGPT, Open LLAMA…) we can see weaker support for Serbian. One of the main reasons for this is the amount of data available online regarding English. One of the possible solutions for filling this gap, of missing data, is digital collections that are available online provided by CHI (Cultural Heritage Institutions). We would see how CHI can enrich the low resource language data, what advantages and disadvantages show in these efforts and how we can solve these problems.

Anton Neznamov, Team Lead, My.Games

A research on automatic segmentation of unknown objects in a batch of videos to reveal which components affect performance the most.

Ana Stojković-Knežević, Project Manager, Protech

Here’s a concise structure for a 10-slide presentation on effectively managing AI projects with a focus on Data Driven Scrum (DDS): Slide 1: Introduction to AI Project Management – Brief overview of AI project management complexities and the necessity for specialized frameworks. Slide 2: Understanding AI Technologies – Importance of grasping AI capabilities and limitations for project planning and execution. Slide 3: The Role of an AI Project Manager – Highlighting the bridge between technical AI aspects and strategic organizational goals. Slide 4: Agile Methodologies in AI – Introduction to Agile and its adaptation for AI through Data Driven Scrum (DDS). Slide 5: Data Driven Scrum (DDS) Overview – Key features of DDS and its advantages in managing AI projects. Slide 6: DDS vs. Traditional Scrum – Differences between DDS and traditional Scrum, emphasizing the flexibility and focus on data and experimentation in DDS. Slide 7: Implementing DDS in AI Projects – Step-by-step guide to applying DDS in AI project management, including iteration planning and execution. Slide 8: Key Roles in DDS – Description of roles within DDS: Product Owner, Process Expert, and DDS Team Members. Slide 9: Challenges and Solutions in AI Project Management – Common AI project challenges and how DDS addresses them, focusing on iterative learning and adaptation. Slide 10: Conclusion and Future Outlook – Summary of the importance of effective AI project management and the evolving role of methodologies like DDS in the face of advancing AI technologies. Each slide should encapsulate the core ideas succinctly, with visuals and bullet points to engage the audience and make the information memorable.

Tamara Zavišić, QA consultant, IWConnect

I would like to present the Ethical AI Development Framework that is applicable not to a particular stage of the AI development process but being of a holistic nature, and is applicable in addressing the entire life-cycle of the AI technologies including the stages from design and development to deployment and in-service as well as post-deployment. The purpose of this presentation is to focus on the social implications of Artificial Intelligence technology, an aspect often less heard of, particularly focusing on ethics. The integration of ethical principles into the development of this technology acts preemptively to ensure its conformity with emerging legal norms.

Tihomir Opačić, Founder, Orange Hill Development

The talk will focus on the past, present, and future of generative AI from a technological standpoint, but with a strong focus on the impact on business, society, and individuals. The talk will first cover critical historical events that led to the current Generative AI development state: – Overview of Ray Kurzweil’s “The Singularity Is Near“ book concepts and why the AGI/singularity predictions still hold true – A brief history of Google’s Deep Mind, major achievements, potential, and AGI as the ultimate goal – Friction between Elon Mask and Larry Page that ultimately led to the Open AI formation – A brief history of Open AI, major achievements, potential, and AGI as their ultimate goal – An overview of notable competitors in the field of Generative AI The talk will then present an overview of the current development and business events. – An Open AI – Microsoft – Google – Meta – X rivalry – An overview of the current level of Generative AI adoption and its impact on business, society, and individuals – An overview of the pace of development in the past year to be able to gauge the direction and speed of future development Finally, the talk will address the predictions for the future, both near and long-term, with a focus on: – Impact of Generative AI on Programming Industry and acceleration of future AI R&D – Impact of Generative AI on traditional businesses, society, and individuals – What businesses, society, and individuals can do to adapt and survive the upcoming changes