• Table of Contents

Introduction

Meepoer is an AI powered social media posts generator. (Named after the infamous Dota2 Hero Meepo).

Check out the repo

https://github.com/TyroneHe-0926/MeepoerPublic

Meepoer was inspired by some of the newer technologies I played around with during my previous work term.

The idea was to utilize GPT-3 (was still in beta) to build a tool for sales and marketing, which transitioned into building a social media post generator late on.

Meepo!

Imagine that you built a great product and now you want to advertise it on different social media platforms, instead of having someone creating 10 different social media accounts to showcase your product, Meepoer can do that for you within seconds. Throughout this post, I wanted to share a little bit about the process of building it, and some key takeaways.

Features

Meepoer currenly supports generating posts for Instagram, Twitter, Reddit, and Facebook. Users could also configure the tone, formalness, writing style and a lot more.

Besides generating content on social media, it could also help writing Youtube captions, blogs, emails, and automate the task by setting a scheduled run.

The overall generation process follows:

• Write a short description on what the post is about, or simply upload an image similar to the post that you want to generate with Meepoer.
• Configure Meepoer’s generation settings including mood, tone, length, and more.
• Now Meepoer will analyze the post and tag it with a category. After that, it will search within our internal post database to find a similar post of the one that you wish to generate.
• After having a few results returned from our internal database, Meepoer will combine the best results and generate a few post based on these best matches.
• Now you can select the post that you think is the best and share it on the social media platform of your choice.

Implementations

Data Sources

Like I have described in the feature overview, having a large amount of data to compare to and fine tune models from MovieNet is the base for all generations. So the first step for building out or infrastructure is to scrape a bunch of data from popular social media platforms. We chose to scrape Instagram, Twitter, and Reddit. Here are the tools we used for scraping the data.

• Twitter: Twitter API
• Reddit: Reddit API
• Instagram: https://github.com/althonos/InstaLooter

We didn’t really wanna pay a lot for s3 storage, and didn’t have any physical storage with us back then, so we stopped with around 110G of Twitter posts, and 115G of Reddit posts, with their respective provided API.