Synergys
Showcasing Advanced Language Technologies in Action
Validation Approach:
Our Eloquence project pilots are foundational to validating the multilingual and multimodal technologies we’re developing. Scheduled to commence in the sixth month, these pilots are orchestrated by leading entities including OM, UNS, TID, and CNR. Each pilot integrates and leverages the innovative tools created across various work packages (WPs), providing crucial insights that influence ongoing refinements. This rigorous validation ensures that our technologies not only meet but anticipate the complex needs of diverse communication scenarios.
Collaborative Refinement:
The pilots are interconnected with all facets of the project—spanning research and development, ethical considerations, and societal impacts. To maintain cohesion and ensure the seamless integration of components, we facilitate bi-monthly meetings with project leaders, developers, and stakeholders. This structured collaboration aids in aligning the technological advancements with our overarching project goals, ensuring that each pilot is optimally positioned to test and demonstrate the project’s achievements in a real-world environment.
Tech Deployment:
An Interactive Playground (IP) will showcase and test linguistic models and tools tailored specifically to our unique use cases. Customized to accurately mirror the unique data and dialogue flows of our partners, these tools are adapted for pilot testing various advanced language model prototypes. Supported by BSC’s robust infrastructure, this setup provides an optimal cloud HPC environment for training, testing, and integration. Feedback from these pilots will refine our models, ensuring their technical prowess and social integrity.
Smart Home Privacy with Decentralized Training
Leveraging Federated Learning, this pilot enhances privacy in smart homes by developing decentralized, personalized language models that secure user data while enabling service improvements across various Telefónica platforms.
Bias Detection in Language Models
Focusing on detecting social biases, the pilot utilizes cultural context to ensure language models in customer service scenarios respect societal norms and contribute to more equitable interactions.
Empathetic Virtual Agents for Customer Service
Enhancing customer service through retrieval-augmented LLMs, the pilot develops virtual agents that understand user goals, make informed API calls, and respond empathetically, aligning with user emotions and objectives.
AI-Enhanced Emergency Call Center Supervision
Transforming emergency health support in call centers, the pilot uses AI to supervise multimodal dialogues, assisting medical staff to provide timely and accurate advice through an innovative audio corpus and AI recommendations.