According to Tensorflow architecture page, the following figure illustrates the general architecture of Tensorflow runtime architecture. A C API separates user level code in different languages from the core runtime.
How does python interpreter fit in the picture? In parallel to Tensorflow runtime?pythontensorflowcpython
Most of the parts above the C API are implemented in Python. You run those parts using a Python interpreter.
More verbosely, the C API and below, is basically just a library. You can use this library from different languages. For training, you usually use Python (all tf.* APIs), which runs using Python interpreter. For inference, you usually use Java or C++. In this case, no Python interpreter is necessary.