I would like to share the project that I've been working on for the past six months. It's called "Sematic", and it's an NLU API that parses plain english so that devs can really start start to understand the language of our users.

In January I was trying to build a "text to bash script" product for my wife, as she is unable to use services like Zapier due to security reasons. I ended up being completely stuck on the first part of the project, which was parsing plain english to a data abstraction that I could code against. The largest problem I had was determining all the relationships between different objects/entities.

I was hoping that Machine Learning could help me, but after going down some rabbit holes, building models, abusing GPT-3, etc, I found that no matter what, I was going to need proper error correction on the model's output. This prompted me to begin studying linguistics, and other no-ML rule based language parsing options. I still feel like there is going to be some ML in this project eventually, but for now I feel like I still have a lot of alpha to go by using some great libs like Duckling, and using old fashioned parsing techniques.

My goal is to get Sematic to a point where devs feel comfortable with using it to add Alexa/Siri-like text understanding to all kinds of apps and websites. If you have comments or suggestions, I'm happy to chat more about it. Thank you.