Skip to content

LLMs in Action: From Vision to Intelligence

Photo of Kostya Kilimnik
Hosted By
Kostya K. and Sergey G.
LLMs in Action: From Vision to Intelligence

Details

Y-DATA Meetup #23
LLMs in Action: From Vision to Intelligence

Powered by Nebius and hosted by AI21.
Talks are in English.

Directions: AI21 offices are located at 14 Leonardo Da Vinci St, Tel Aviv. Offices building B, entrance 12A, 4th floor

* More info about new course by Y-DATA.

Agenda:
18:30 - 19:00 Registration, Mingling, Snacks & Beer
19:00 - 19:30 Lightning talks
Andrey Nikitin (Cyera) - No Limits: The File Level Model That Classifies Any Document
Urska Jelercic (Lightricks) - Beyond the Pixel: Identity Preservation in AI-Powered Image Editing
Udi Cohen (Vendict) - AI-Native Trust Building
19:30 - 20:00 Keynote by Gal Chechik (Nvidia) - Recent advancements in Visual Gen AI
20:15 - 21:00 Panel Discussion "The Future of Agents" with
Amit Mandelbaum (AI21), Uri Goren (Argmax), Shir Chorev (Deepchecks), Simon Karasik(Nebius), Serj Smorodinsky (Loris)

Abstracts:
"Recent advancements in Visual Gen AI"
Between training and inference, lies a growing class of AI problems that involve fast optimization of a pre-trained model for a specific inference task. These are not pure “feed-forward” inference problems applied to a pre-trained model, because they involve some non-trivial inference-time optimization beyond what the model was trained for; neither are they training problems, because they focus on a specific input. These compute-heavy inference workflows raise new challenges in machine learning and open opportunities for new types of user experiences and use cases. In this talk, I describe two main flavors of the new workflows in the context of text-to-image generative models: few-shot fine-tuning and inference-time optimization. I'll cover personalization of vision-language models using textual-inversion techniques, and techniques for model inversion, prompt-to-image alignment and consistent generation. I will also discuss the generation of rare classes, and future directions.

"No Limits: The File Level Model That Classifies Any Document"
In this talk, we introduce the File Level Model, a novel solution for document classification that processes files such as Word documents and PDFs and assigns meaningful classifications—even for document types it has never encountered before. Traditional multi-class models are constrained by fixed class labels and can only recognize categories they have been explicitly trained on. By contrast, our File Level Model leverages a fine-tuned LLM, enabling it to adapt to a vast and evolving range of document types. We will discuss how this approach overcomes the limitations of predefined class sets, and showcase its real-world applicability in classifying new or specialized documents.

"Beyond the Pixel: Identity Preservation in AI-Powered Image Editing"
Inpainting, retouching, and content-aware edits powered by generative AI require more than just realism—they demand strict identity preservation. This talk will discuss recent work done at Lightricks Facetune research team that ensure AI-driven modifications remain faithful to original content, balancing flexibility with fidelity in creative applications.

"Vendict - AI-Native Trust Building"
Companies rely on vendors to drive innovation, but how can they be sure those vendors are secure and follow all the rules and best practices needed to protect their business and reputation?
Vendict makes collaboration easier by providing instant “trust but verify” assessments using specialized AI technology.
Our first product was built entirely on proprietary models. As LLMs and AI agents continue to advance, we continue to develop and integrate proprietary models at critical points in our pipeline, ensuring a competitive edge and a solution that truly works.

Photo of Y-DATA meetups group
Y-DATA meetups
See more events
AI21
Leonardo da Vinci St 14, 4th floor · Tel Aviv-Yafo