Based on my brief browsing of the code, it looks like it's all in the browser cache itself. The bytes are split into numbered chunks, converted to b64, and then a sequence of QR codes are generated from the b64. At the end the received data is crc32 checked for validation. There are adjustable parameters and a progress bar, making it easy to retransmit any chunk that wasn't properly received.

The code is incredibly easy to read, everything is in a single HTML file with zero obfuscation (unless you count the two minified QR code dependencies that also include links to the un-minified versions).