Comet’s annual Convergence conference gathers AI and data-science practitioners around the latest industry trends. The Convergence is one of Data Science Salon’s partner events, where professionals can exchange thoughts and remarks in a welcoming environment, and listen to expert talks on current AI trends.
The event spans two days, packed with knowledge and fresh ideas. With 25+ speakers and more than 3,500 attendees, the conference is a perfect opportunity to exchange thoughts, opinions, and remarks on the current state of AI.
The event is held online, so participants and enthusiasts from all over the world can enjoy the knowledge shared.
Surfing the trends
Delivered “by practitioners, for practitioners,” the conference features talks that explore the latest technologies shaping modern AI and machine learning. Each “trend” is more than a single topic—it’s the intersection of multiple aspects of a solution, technology, or phenomenon. This edition explores the following trends:
AI agents
AI agents are attracting growing attention from businesses and data scientists worldwide. As autonomous as possible, these agents perform tasks within a company’s systems with little to no human supervision. They hold impressive automation potential, from running marketing campaigns across ad platforms to supporting sales teams or helping IT specialists maintain systems.
AI agents will be the focus of the panel “The Rise of AI Agents: From Demos to Deployment,” where experts from Comet, Meta, Pattern, and CrewAI will share their knowledge and ideas.
Alex Shershebnev, ML/DevOps Lead at Zencoder, will present “AI Coding Agents and How to Code Them,” detailing how to automate coding tasks and wisely apply these tools day-to-day.
“We’re talking about a tireless, always-enthusiastic partner ready to help. At the same time, it is not a replacement for the developers (yet!), as it can misinterpret requests and do pointless things. Overall, I’m highly enthusiastic about coding with AI agents, and there’s far more to it than just copy-pasting code from ChatGPT,” he says.
Context retrieval and hallucinations
Generative AI brings inspiring possibilities—but also new risks. AI hallucinations, where a system fabricates plausible-sounding information, are among the biggest. Without tackling them, production deployments become risky.
“Would anyone trust a calculator that answers 2 + 2 correctly with 4 most of the time, but sometimes it answers 5 and occasionally ‘tomato soup’? I doubt it. Yet that’s how hallucinations work, and thousands trust these systems without fully understanding the risk,” says Jasleen Singh, Senior Principal Software Engineer at Dell. She’ll address the challenge in “From Hallucination to Accuracy: Evaluating Context Retrieval in RAG Systems.”
Optimizing LLM and AI performance
Large Language Models are—by definition—large, sometimes gargantuan, with billions of parameters.
“But bigger isn’t always better,” says Rajarshi Tarafdar, Senior Software Engineer at JPMorgan Chase. “DeepSeek showed how smaller, less-expensive models can be both cost-efficient and higher-performing. Overall performance is a far more complex matter.” He’ll dive deeper in his session, “Optimizing LLM Performance: Scaling Strategies for Efficient Model Deployment.”
AI safety
AI systems increasingly shape our lives—from spam filters that might hide a vital email to diagnostic aids that can miss a disease or banking models that tweak interest rates based on obscure criteria.
“Making these systems safe and reliable is a big topic the industry must address,” says Allegra Guinan, Co-Founder & CTO at Lumiera. Talks such as “Evaluation-Driven Development: Building Reliable AI Systems” and “Practical Approaches to Building Safer AI Systems” will offer insight and inspiration.
AI sustainability
The International Energy Agency estimates that data centers, AI, and cryptocurrencies consume about 2 % of global electricity demand (460 TWh).
“Reducing energy use tackles climate change and cuts costs,” notes Ram Ganesh, Senior Data Scientist at Mastercard. “Ensuring AI is truly accessible to all requires us to keep sustainability at the forefront. Building efficient models isn’t just an industry requirement- it’s a vital step toward a scalable, inclusive and environmentally responsible technology”
He’ll explore this topic in “Smaller, Smarter, Sustainable: Optimizing LLMs by Leveraging White-Box Knowledge Distillation Algorithms.”
Summary
Alongside expert talks, Convergence features panels where practitioners discuss the most urgent challenges and opportunities the industry faces.The conference runs online May 13–14. The full schedule and speaker list are available on the event website.
