5 years ago, before covid, I created my first bot.
The bot was replying to any mention of me on twitter by a built up quote of Rick from Rick and Morty.
It was not an LLM. It was built using an
n-gram algorithm fed with all the script of the tv-show.
As I've recently lost my job, gaining therefore some free time, I decided to start other projects.
I was told that if I wanted to market my projects for free, I needed to build a personal brand on Linkedin,
which meant posting regularly on the platform. I got bored by my second post.
So I thought I'd get the help of the LLMs to repeat more or less what I did those 5 years ago.
I created a new Ruby On Rails app ( a framework based on the language ruby which I love).
I then went to read the documentation on what Linkedin needed for me to communicate with its API.
( an API is what is needed to programmatically interact with an application ).
It needed me to register a company, which I did. And then to have users login through my application so they can
give consent of having posts created from the app.
I had to go through the devise oauth documentation again
( devise is a user management library in ruby and
oauth is an access delegation standard - when you don't want to
sign-in with your email but with your social media account, it's most of the time an oauth system)
Once I could login through linkedin with my app, I had to communicate with an other api: a decent LLM.
I decided to use Claude from anthropic, as I had already some fun with ChatGPT and wanted to change scenery )
Well, there isn't much change in scenery as they use the same structure of communication.
I gave Claude a description of myself, and then gave it a prompt, like "write me a linkedin post".
It then posted it.
First added value of the app: check.