How I built and sold my startup to the largest mobile game company in the world

School of Computing graduate Sebastien Marion will be returning to Kent to share his story of the ups and downs of his start-up success.

His talk ‘The startup journey, how I built and sold my startup to the largest mobile game company in the world’ will be in Cornwallis SW101, 15.00, Tuesday 29 November and is open to everyone.

After completing his PhD in 2008, Sebastien went on to create Comufy, the world’s most scalable customer relationship management (CRM) platform. Comufy was eventually acquired by King Digital, the largest mobile game company in the world and founder of Candy Crush, in 2014 to power their game-to-player communications.

Sebastien is also investor, advisor and board member at Yugo, the electric scooter sharing service.

Seb completed his PhD in Computer Science  under the supervision of Professor Richard Jones. His dissertation was entitled  “Using Class-Level Static Properties to Predict Object Lifetimes”. Modern computer applications allocate memory (objects) at a prodigious rate, many megabytes per second. In order to prevent the computer from running out of memory, it is essential to recycle an object once the program has finished with it.

This turns out to be an extremely error prone task for the human programmer so modern programming languages use an automatic technique called garbage collection to recycle unused memory. A naive garbage collector might repeatedly check whether each object is still in use but this is inefficient. A more sophisticated generational garbage collector takes advantage of the observation that “most objects die young” to segregate young objects (which are examined often) from old objects (that are examined only occasionally).

The innovation in Seb’s work was to use machine learning to predict which places in a program would allocate short-lived objects, which would allocate long-lived objects and which would allocate objects that remain in use forever. Once Seb’s system had learnt from a few training programs, it was then able to predict quickly and accurately the object-lifetime behaviour or programs that it had not seen before.