In a world where the demand for generative artificial intelligence is increasing, particularly in the business sector, innovative techniques such as retrieval-augmented generation (RAG) and fine-tuning are emerging. These methodologies aim at the specialisation of foundation language models, but we still face several challenges, such as the integration of external data, relevant context retrieval and the reliability of responses or hallucinations.

The purpose of the session is to present examples of how these problems are addressed through real, operational use cases with a multi-sector scope.

Presents:

• Barbara Eisele, AI Community, Digital Catalonia Alliance, i2CAT foundation

Participants:

• Sanket Biswas, Ph. D candidate, CVC

• Daniel Font de Matas, CEO, HaltoBooks S.L.

• Fernando Bayón, CTO, dezzai

www.cvc.uab.cat | CVC

Sanket Biswas is a Ph.D. candidate at the Document Intelligence research unit in the Computer Vision Center (UAB). Since earning his Master’s in Computer Vision in 2020 from UAB, his research has focused on multimodal document generation, document layout understanding, scene text recognition, and sketch generation. Sanket has co-authored over 20 papers presented at top conferences like AAAI, ICRA, ECCV, CVPR, WACV, and ICDAR. He has further honed his skills as a Research Scientist/Engineer intern at Adobe Research’s Document Intelligence Lab. Known for his significant contributions in Document AI, Sanket has appeared as an invited speaker at workshops organized by tech giants like Huawei and ServiceNow.

This talk introduces the “DocSynth” series, a generative approach in the realm of multimodal document generation which combines both layout and content (both visual and textual). Our models utilize an autoregressive structured language model to seamlessly integrate both layout and textual cues to enhance the generation of contextually relevant and structurally coherent documents. We also curated a new benchmark, showcasing the model’s capabilities in producing high-quality documents without relying on visual components. This advancement not only elevates document generation techniques but also opens new avenues for research in automated document and layout design.

www.hal2b.com | Linkedin

Daniel Font de Matas has a degree in Business Administration (ADE) and specializes in Strategic Marketing, as well as being passionate about new technologies. With more than 20 years of experience in Marketing and Product areas (B2B and B2C) in companies such as Grupo CIRSA, Lonely Planet, or Grupo Planeta, he has led teams in international projects in South Korea, Venezuela, and the USA. His passion for innovation and his recent entrepreneurial approach have made him an expert, highlighting his ability to generate impact and provide solutions that drive the success of companies in the publishing industry. In 2022, still without the media noise of AI, he launches Hal2Books (hal2b.com), a tool for the sector, which analyzes and creates reading reports from manuscripts.

The industry of content, and especially the publishing industry, is currently where the culture of self-publishing and the generation of new digital tools are causing interesting movements in a sector with few changes since the age of printing. Due to our wide experience, we see that there is currently a real need to identify new publishing bestsellers and to have tools that help authors improve their works. In this sense, our proposal is SALANDER (hal2b.com), a tool based on Natural Language Processing that allows publishers to find new talent among the enormous mass of self-published authors, and authors to improve their own works.

dezzai.com | Linkedin

Fernando holds a degree in Computer Engineering from the Autonomous University of Madrid (UAM), Executive MBA from the School of Industrial Organization (EOI) and EXMBA from the University of Central Florida (UCF). He has more than 20 years of experience leading technology teams and 10 years implementing Artificial Intelligence in different industries, banking, healthcare, utilities, fintech.
Prior to dezzai he was CTO of Repsol’s Waylet and CTO of Sanitas Bupa, building high performing technology teams and applying technology in adapting business units to the digital economy. Since 2018, at dezzai, he leads technology teams and designs R&D strategy in Artificial Intelligence to generate competitive advantages in various industries.

A digital assistant for the efficient interpretation and application of occupational health and safety regulations will be presented. This resource, key to legal compliance and the promotion of safe work environments, will provide accurate guidance and immediate access to crucial information. It will be highlighted that this type of virtual assistant can be tailored to meet the needs of all industries, ensuring its widespread utility in a variety of work contexts.