Here I tried to cover most common scenarios. There may be lot of other combinations of input file. Anyhow what you are saying is correct, and we can use 'from_json' instead of 'json_tuple' with dynamic schema creation based on the input types (row wise)