The challenge: evaluating the potential of generative AI in production
The project aimed to determine to what extent generative AI could be reliably integrated into a weather-related audiovisual production environment.
The project needed to evaluate the quality of the generated content, its suitability for editorial and audiovisual constraints, as well as the performance, costs, and technical limitations of the various available models.
The prototype also had to combine text generation, graphic design, cloud infrastructure, and workflow automation within an architecture suitable for industrial-scale deployment.
Project management: turning testing into a prototype
Aski-da Synergies oversaw the scope of the study, defined the target workflow, and coordinated the expertise brought in from the fields of art direction, graphic design, software development, and graphic engineering.
An initial analysis phase mapped out the production cycle of a newsletter and identified the main opportunities for automation.
A benchmark of generative AI models was then conducted to compare their linguistic quality, performance, usage costs, and relevance to the use case. Through prompt engineering and experimentation, the generation of bulletins was gradually refined.
The project was carried out in partnership with AWS for the cloud infrastructure and Engine for the graphics and real-time components. Several validation stages enabled the project to evolve gradually, from a text-based POC to a graphical prototype and then to the final prototype.
Results
The project validated the feasibility of a weather production workflow incorporating generative AI and identified the conditions necessary for its industrialization.
A functional prototype capable of automatically generating a complete weather report was delivered, along with documentation of the workflow and recommendations for the next steps.
