Deeptea - NLP, Deep Learning Simplified

Published Aug 22, 2017

About me

I'm a software developer that loves building creative, beautiful, and useful products and solutions. I work on both proprietary and open source software, with over 10 years of combined experience.

The problem I wanted to solve

Natural Language Processing and Deep Learning is complex - especially if you don't understand it, or can't wrap your head around it -- I was like that for a long time, then it finally clicked, it was so much simpler than I imagined! So, I made it twice as easy for anyone to understand, with Deeptea - a deep learning-capable, natural language processing tool.

What is Deeptea?

Deeptea is a smart library for building deep learning-capable, natural language processing applications.

Tech stack

ECMAScript 6

The process of building Deeptea

First, I learned about how natural language processing and deep learning works (to an extent, I'm still figuring it out even today!), then I used some existing modules to start building my library, and adding more powerful functionality, and giving usable output, instead of garbage output. We do all the processing and handling on the Deeptea library itself, you just receive an output. We wanted to make training our model (AI) dead simple, so we wrote a teach wrapper that would allow you to write case-specific rules for inputs, outputs, etc.

In the end, we're left with a dozen of complex features, made dead-simple, in under 7 lines you can have your own deep learning AI model and processing of natural language, and build pretty awesome apps.

Challenges I faced

Learning how deep learning works was a struggle - I'm no good at math, but I figured it out after 6 months of trying to learn it. The big challenge I faced once I figured this out, was practically applying it, which took 3 months to design how it should feel to users.

Key learnings

Natural Language and Deep Learning is complex, difficult, and extremely long to do - if you don't know how. However, with Deeptea, it's now simplest it can be, and it's still growing. Further aspirations for deeptea include is dynamic training based off of user input, automated data scanning (all your tweets, etc. being classified and used as possible replies? yep!).

Tips and advice

Learn NLP and Deep Learning with Deeptea - it gives you a easy, higher-level way to dive into this complex world, and give your application intelligence.

Final thoughts and next steps

We're still building beautiful applications, and it's growing daily - all thanks to deeptea. If I could correct one thing, it would be how the original API worked for deeptea, it was in-and-of-itself complex, but as of v0.0.2+ this is fixed, and now it's as simple as pie.

Discover and read more posts from Curtis Gervais
get started