This is under construction.

Currently, I have a Python web server that, given a user query, performs a semantic search (basically using AI transformers to vectorise inputs) to find the most semantically-similar newspaper headlines from a dataset of 1950-1999 American regional newspaper headlines.

The Python code is very CPU- and memory-intensive, thus it would basically DOS my server if I allowed people to use it.

If I rewrite it in C++, I could compile it to WASM, thereby transferring the CPU-intensive parts to the viewer. But this would leave me with a huge bandwidth problem, because viewers would have to download gigabytes of vector data before using it.

Source code, link to the dataset, and very limited demonstration will be available some day, pending this rewrite.