So, investing in AI is a large risk: back to my university, the popular AI framework is a neural network, logistic regression, supported-vector machines, etc.
Instead, I spent time working on data infrastructure, which keeps evolving but follows a somewhat consistent direction. If you look at databases, the essential building blocks are identical to those fifty years ago.
I first worked on Apache Flink and became a committer after one year as a contributor. And then Apache ZooKeeper and Curator, so did become a committer / PMC member. Later TiDB / TiKV and now GreptimeDB. The knowledge and experience is reusable.