Mappable is hopefully a new way to build text classifiers (and eventually other NLP models).
Here is an example project that you can explore (or help annotate!)
https://mappable.ai/5GyAj6CKnk2E8x67SjgcpF
Mappable is based on a few ideas:
Search as Annotation: Programmatic labelling is great for technical users. But search is a pretty good alternative, with a UX that is familiar to a lot more people.
UMAP/TSNE Visualisation: In the graph tab of a project, you can annotate data by selecting data visually. This can make the process a lot faster if you have coherent clusters in your data.
Data Collaboration: We collaborate on code so easily with github - why can't we do the same for datasets? In Mappable, you can annotate any public project, which will be submitted to the owner as a "pull request", which they can accept or reject.
Models you train in the Mappable web interface are automatically deployed with a small interactive demo, and an API for downstream use.
I am at a bit of a loss where to take this project, so if you have feedback I would be very interested to hear it (technical or business wise!).
