TANGO Architecture

The TANGO Ecosystem is the technological backbone of the project, designed through a co-design approach where scientific methods, enabling technology and high-impact case studies are developed in tandem. The architecture is composed of three primary outputs: the TANGO Library, the TANGO Models and the TANGO API. The foundation is the TANGO Library, an open-source (Apache v2) Python library that implements all the novel methods for synergistic human-machine learning and hybrid decision-making developed within the project. This library serves as the core toolkit for researchers and developers to build, train, and refine advanced XAI and HDSS models.

Once trained, these TANGO Models are deployed to a cloud infrastructure and are exposed through the multi-purpose TANGO API. This API is a critical component, enabling application developers to seamlessly integrate the sophisticated capabilities of TANGO's research into their own products and services. Its design is inspired by the structure and usability of the OpenAI API, ensuring ease of use for innovators.

The internal structure of the TANGO API is modular and robust, designed for scalability and security. An API Manager serves as the entry point, handling incoming requests from the various case study applications (e.g., perinatal well-being app, credit lending app). These requests are then passed to an Orchestrator, which coordinates the various backend services. An Authentication/Authorization Server ensures secure access to the system's resources. The core logic is handled by specialized engines, including an Explainability engine, an Interaction dialogue engine, and the underlying ML models, all of which access a shared Knowledge Base. All interactions are logged in a dedicated storage component, which feeds into Monitoring and Analytics systems to track performance and usage. This entire ecosystem is developed using an agile methodology with continuous integration and deployment (CI/CD) pipelines. Crucially, cybersecurity and ethical considerations are built-in, ensuring that any personal or sensitive data processed by the system is protected and handled in strict compliance with GDPR and the principles of Trustworthy AI.