Our initial approach was to store the full board in a single row in Cassandra and each request for the full board would read that entire row.
This is the epitome of an anti-pattern .I sincerely hope that this approach was floated by somebody who had never used Cassandra before.
Even if individual requests were reasonably fast, you are sticking all of your data in a single partition, creating the hottest of hot spots and failing to leverage the scale out nature of your database.