Introduction

The first ever AI project that I ever made was a persona based AI project where I trained a GPT-model Gemini-2.0-flash and it was successfully able to mimic the behavior of famous tech entrepreneur Hitesh Choudhary. It was not as much of a rocket science as I believed it to be. In this article I will explain how I created the chatbot without making frontend and how its only one prompt away from becoming your own custom persona.

Tech Stack Used

I wanted this project to be a simple beginner friendly tutorial. Therefore, I didn’t make frontend myself, rather I used Streamlit a free frontend builder.

Gemini API - Free APIs available, you can also use OpenAI’s API as it is the industry standard, however no free options are available, a minimum recharge of 5$ is sufficient for this and many projects
Python >= 3.13 - I coded this in popular ML language python
uv package manager - I am not a fan of how we have to set up virtual environment and make a requirements.txt file every time we make a project, so instead I used a fast python package called uv by Astral docs. It creates a project environment and “it’s essentially like npm for python”
Dependencies - python-dotenv, google-genai, pillow and streamlit

Walkthrough

First I initialized the project using uv init. This will generate the pyproject.toml file and initialise a virtual environment for you.
Then in main.py, I imported all necessary modules and made the GPT client. Then I learned how to use streamlit through their docs and made an average UI
I made sure to store my api key in a .env file as it is a safe measure and standard
Then, the most important part of the project, the SYSTEM_PROMPT. This is the part which is known as prompt engineering generally. This is where yu set the rules and explicitly mention how the AI should behave. I inserted data of my desired persona here for it to mimic that person

Problems I faced and how I solved them

During this project I faced many errors but with each error comes a new opportunity to learn
Chat history was one of the things I struggled with, however with the help of docs I learned about streamlits session handling capabilities. With my newfound knowledge I was able to implement a fully functioning chat history after some tweaks
I made some basic errors in my project and solving them clarified my python project boilerplate understanding
One of my biggest mistake was installing dotenv when the correct dependency was python-dotenv
Also after deployment my app did not run, after some troubleshooting I realised that streamlit cannot access the .env file so it could not detect the API Key. So it needs to be given in a special file that can be accessed in streamlit cloud

Takeaways

You do not need mathematics for making chatbots for development purposes. They are needed by data scientists
By altering the SYSTEM_PROMPT anyone can create any persona
Not knowing how to make UI is not a barrier towards positive ML development
We can only make a project through problem solving and debugging
Anyone can create chatbots with just a little bit knowledge about GPT. To learn more see my article on GPT for dummies