An AI concept design for real-time highlight generation





Empowering industries with customizable AI models

This concept design was inspired by our experience at Evolphin, where we saw how artificial intelligence (AI) could revolutionize various industries. One of our projects involved developing an AI system that could automatically identify critical moments in a football match and generate pre-edited clips for editors to refine and publish in real-time. This system was used to create highlight reels of the match, and was also utilized by the sales team to provide popular brands with a summary of where their brand appeared and with whom.



However, we encountered a challenge in making this AI system repeatable and adaptable to different contexts. To overcome this obstacle, we created a concept platform that enables clients to create and fine-tune their own AI models according to their specific project requirements. This platform makes the power of AI more accessible and customizable to diverse industries and use cases.





Developer-friendly ai experimentation with low-code environment

The platform was designed for developers and provided a user-friendly environment for experimenting with AI models in a low-code or no-code environment, similar to Stripe's approach in finance. Once the training was completed, users could call their models via an API using the platform.



In the future, the platform could offer pre-populated templates or HTML building blocks for a more low-code or no-code approach.





Streamlining ai testing with reusable templates and datasets

The tool allowed developers to create test cases directly within it, and switch between different datasets and templates to test the AI model more thoroughly and ensure it performed as intended.



By utilizing templates and datasets, developers could automate this process and make use of reusable building blocks.







Flexible ai training cycles for low-cost concept testing

Developers could create their own training cycles that could stop based on a cost cap, number of epochs done or until a certain level of similarity was achieved.



This full control over the training process helped engineers and product managers test concepts in a low effort and low cost environment to test their ideas.





Efficient ai dataset management with manual editing and vectorization

The platform provided developers with a convenient way to manually edit imported datasets, enabling them to correct any mistakes and improve the dataset's quality by adding more content.



Additionally, the platform offered the option to vectorize large datasets to optimize them for AI training. This feature allowed developers to work more efficiently with large amounts of data and enhance their AI models with greater accuracy.







Get alerts when your ai tasks are completed

To further enhance the user experience, we implemented an early notification system to alert users when certain tasks were completed. Given that both training and testing could take a considerable amount of time, we considered this an essential element of the platform's user interface.

This feature allowed users to stay up to date with the progress of their tasks and helped them manage their time more effectively.