Contact
LOG IN
Sign up

Oct 30-31, 2025
Take your AI agents from prototype to production

0
Days
0
Sessions
0
Speakers
0
Attendees
OVERVIEW

What is Fully Connected?

Fully Connected is your chance to join top AI engineers for two days of agent building and model training. We'll help you reach the next level in AI development, with experts from everywhere, hosted by Weights & Biases. Day 1 is hands-on workshops cutting across all domains of agent construction. Day 2 features talks by industry pioneers, sharing hard-earned insights across verticals and applications. You'll walk away with a whole new toolkit for building AI applications that matter.

On-Demand

Watch the Sessions

Catch up on all the sessions from Fully Connected

Fully Connected Tokyo 2025: Opening Keynote with W&B Cofounders Lukas Biewald & Chris Van Pelt

This keynote reflects on the rapid acceleration of AI in 2025, as models scale exponentially and real-world adoption spreads across healthcare, law, development, and the enterprise. With benchmarks becoming saturated and agent workflows emerging, AI is fundamentally reshaping how software is built and used. The talk also highlights new challenges in reliability and production, and introduces how Weights & Biases and CoreWeave are enabling the next generation of model development and deployment at scale.

How to Build AI Agents Capable of Competing in Real-World Settings

How can we create AI agents that achieve results comparable to human experts? AI agents developed by Sakana AI, such as “AI Scientist” and “ALE-Agent,” generate creative papers and algorithms. When evaluated under the same conditions as humans, they receive objective assessments that place them on par with specialists. This presentation will explain the technologies underpinning these cutting-edge AI agents, including scaling during inference.

Mistral AI Ecosystem for Japanese builders and enterprises

In this talk Joffrey presents Mistral AI ecosystem, and Mistral AI's vision and values. He then covers their latest releases and how Japanese builders and companies can leverage them though live demos.

Trends and Outlook for the Ministry of Economy, Trade and Industry's AI Policy

Amid the advancing AI revolution, the Ministry of Economy, Trade and Industry launched the “GENIAC” project in February last year. Through initiatives such as computational resource procurement support, promoting data utilization, and community activities, it is working to enhance Japan's generative AI development capabilities and advance their societal implementation. This presentation will introduce the background, content, and future outlook of these policies.

Evolving Large Language Model Evaluation: Practices and Insights from the Swallow Project

In the research and development of large language models (LLMs), evaluation is essential for accurately understanding their capabilities and limitations. With the recent advancements in LLMs, their evaluation benchmarks and methods also require updates. This presentation will organize evaluation challenges and the latest trends from multifaceted perspectives, including knowledge, reasoning, multilingual support, increasing difficulty, and LLM agents. Furthermore, as an evaluation practice in Japanese LLM development, we will introduce the evaluation frameworks swallow-evaluation and swallow-evaluation-instruct developed within the Swallow Project.

LLM Research and Development Initiatives at the National Institute of Informatics

The National Institute of Informatics established the Large Language Model Research and Development Center in April last year. We are leading the llm-jp (LLM Study Group) activities, which have been ongoing since 2023, aiming to build LLMs with high performance on Japanese tasks and enhanced transparency and reliability. This presentation will introduce specific examples, focusing on models we have released so far, including corpus construction for training, training from scratch, tuning and evaluation, and multimodal capabilities.

NVIDIA's Generative AI Ecosystem Supporting Japan's Sovereign AI

Sovereign AI refers to the concept of a nation's ability to create artificial intelligence using its own infrastructure, data, workforce, and business networks. Sovereign AI includes sovereign foundational models enhanced to address specific dialects, laws, cultures, and customs. This presentation will introduce some of NVIDIA's activities related to the growing Sovereign AI movement in Japan and outline NVIDIA's generative AI ecosystem, which offers significant benefits for developing sovereign foundational models.

NHK Broadcasting Technology Research Laboratory's AI Technology Development Initiatives

NHK produces a wide range of content, including dramas, news, educational programs, and documentaries, and there is a growing need to streamline the production of this content. The NHK Broadcasting Technology Research Laboratory is engaged in research and development of AI technologies capable of processing diverse modalities such as text, images, audio, and sign language, aiming to enhance content production efficiency. Furthermore, to enhance safety in operational use within broadcasting stations, we are also advancing the development of our own large-scale language models. This presentation will introduce our institute's AI technology development initiatives, including large-scale language model development, and their application in broadcasting operations.

Post-training best-in-class models in 2025

Overview of state-of-the-art post-training, including data generation, training algorithms, and evaluations.

AI that explains the market: A new paradigm in financial forecasting

In this session, we present EXAONE-BI (Business Intelligence), an AI-powered multi-agent framework for explainable forecasting and equity research. The system orchestrates specialized agents that collect diverse data, identify key drivers, evaluate scenarios, and generate human-readable reports. By integrating external signals such as news, corporate filings, and macroeconomic indicators with internal datasets, EXAONE-BI delivers transparent and actionable insights for equity markets. We will share our collaboration with LSEG as a real-world case study, highlighting how EXAONE-BI supports forecasting, factor attribution, and report generation. Our workflow was further enhanced by Weights & Biases (W&B) for efficient experiment management.

Weaving Together Field, Technology, and Culture: TOPPAN's LLM/VLM Development and Operational Practice

At TOPPAN, alongside AI implementation aimed at business transformation, we are also engaged in the development of LLM and VLM. On-premises AI models are essential for meeting strict security requirements and diverse operational needs. This presentation will introduce, from a developer's perspective, the unique characteristics of each model revealed through evaluation and operation on W&B, as well as models suitable for AI agents. Based on real-world field experience, we will share the current state and future potential of generative AI development.

Introducing Our Approach to Design Document Review Using Business-Specific Large Language Models

Hitachi's Digital Systems & Services Sector is promoting the application of generative AI with the goal of improving system development productivity by 30%. This seminar will introduce initiatives in the Financial Business Unit utilizing a design document-specialized LLM for design document reviews, as well as the use of business-specialized LLMs in WandB for APT/SFT.

The Cutting Edge of Generative AI Adoption in Manufacturing and Toward Building an Air Conditioning-Specific LLM

While the general knowledge and reasoning capabilities of general-purpose large language models, exemplified by ChatGPT, continue to improve, their performance in tasks requiring specialized knowledge remains insufficient. For industrial applications of generative AI, the key challenge moving forward is how to teach these models industry-specific knowledge and internal company information. This presentation will introduce examples of generative AI utilization for design and development within Daikin, followed by a discussion on the necessity of continuous pre-training for an air conditioning-specialized LLM and its development strategy.

Special Discussion Session: The Present and Future of Generative AI

This special discussion between Mr Fukatsu of THE GUILD and Mr Yamamoto of W&B explores how the rapid advancement of generative AI is reshaping business, organizations, and competitive dynamics. As capital becomes directly convertible into labor and development capacity through AI, the session examines what truly constitutes sustainable competitive advantage in an era where tools are increasingly commoditized. Drawing on examples rooted in Japanese industries and culture, the conversation highlights how non-replicable assets—such as history, domain expertise, and structural positioning—will define success in the future of generative AI. It offers a strategic perspective on how companies and individuals can position themselves for the next phase of the generative AI era.

SPONSORS

Thanks to our sponsors