Making an AI personal assistant with zero experience
Very recently I had this concept brew in my head, about an AI assistant that you can give basic instructions through telegram. Of course I had absolutely no idea at that time of how I was gonna achieve this; but I told myself a few Youtube tutorials would do the trick.
Fast forward a couple of weeks of research and I've just began the first few attempts at the project. Below are the things I've learned so far.
How telegram bots work:
I chose telegram for this project as I've seen people do some pretty wonderful things using the telegram bot API and python.
Making the telegram bot was levels easier than I first anticipated. All I had to do was send a couple of commands to an existing telegram bot; Botfather.
From there I was able to pick up the token for my bot; ALYA (stands for Artificial Language Yapping Agent)! Afterwards I looked up a few tutorial articles and followed one by freecodecamp on the basic set up of the bot using telebot; which is the telegram bot api library for python.
So far I've only learned how to make the bot respond to some basic commands such as "Hello" and "Start". Any other message just makes the bot echo it. The telebot documentation has other helpful use-cases which I'm gonna study and try my hand at implementing.
Selecting the proper AI:
I wanted to go with an open source option for this project, of which, there were many to choose from.
My first option was Rasa, an open source LLM that can run on your device locally. However, I came to realize that Rasa was mostly designed to handle instructions and perform specific tasks rather than act as a normal AI chat bot. If I was gonna have an AI assistant, I needed to be able to have atleast a basic conversation with it.
After looking around a bit more I came across GPT4All. GPT4All was an LLM chat application that allowed you to download multiple different language models. And just like Rasa, GPT4All had a python library that I could make use of for the project.
Currently I'm doing everything on a budget laptop; which only has 8GB ram. Due to this I was very limited on which model I could actually use for the project. That being said, GPT4All provides a vast number of models suitable for different situations. I opted to go with the open source Phi-3 Mini Instruct developed by Microsoft.
Future plans:
I've only just started work on the project so for now my goal is just to be able to converse with GPT4All through my ALYA, and get it to do some very basic tasks such as set reminders.
In the future, however, I plan to use ALYA to organize my files, update important information on my devices, and more. I also plan to use ALYA to make my own file system which will make use of tags so that it will be easier for ALYA to pull up relevant files in accordance to my quires!
I'll end by reiterating the fact that I really do not know what I am talking about. Nobody should be using this as a guide; instead look at this as inspiration to do your own research and start your own projects... If you are interested in that sort of thing of course.
Take care <3 -Cakes