Matchmaking System and AI

While at Motiga I revamped their matchmaking system, adopting and tweaking a belief system based matchmaking algorithm to accommodate their needs, there was also a machine learning component that would analyze players performance in matches with different heroes to attempt to better place them into skill brackets.

A simulated or historical set of match made players could be rerun through the matchmaker and analyzed in an admin tool and there were many configurable portions to the heuristic / matchmaking logic. At the core, this matchmaker used a variant of trueskill.

The matchmaking system was decentralized and auto scaled location and skill based pools.

The machine learning system used Kafka and, scala, flume, redshift, redis and spark to analyze players performance in matches and attempt to correlate and classify them continuously into more accurate skill brackets that the matchmaker would incorporate into it's match heuristic. Development on this was spec'd out when I left the company.

Employer: Motiga