So, I’ve been trying to build a chatbot for the last three months. I scraped my friend’s website (with his permission, of course), and converted his content into vector embeddings. Then, I tried to build a chatbot using the Gemini model on top of that.

The first post on the WordPress website was simply “Hello World!” — and my friend’s site was still under development.

I was excited to see how this chatbot would work. I started a chat with it, thinking, “Man, I’ve finally got the vector embedding down! Now the LLM will give me some awesome answers!” So, I typed “Hello” into the chatbot… and you know what it responded with? “Hello World!” 😂 I had literally spent four hours coding, fixing bugs, and working through errors… and LLM just hit me with “Hello World?” Seriously?

Obviously, the problem wasn’t with the LLM — it was in my code and logic. There was no context being provided, and the chunks in the vector embedding weren’t set up properly. At that point, I realized I was completely new to this field. I didn’t fully understand how LLMs worked or how everything should be structured. I just knew the basics from YouTube videos and blogs — what the key concepts were in building AI chatbots.

So, I decided to take a step back, go back to square one, and start learning Python. At the time, I wasn’t even coding in Python — I mainly worked in PHP for my job. But I knew I had to learn it to move forward. After diving into Python, I spent a week learning linear algebra and grasping the basic concepts, knowing that this knowledge would be foundational for AI and ML.

Then, something interesting happened. My cousin, who’s been into AI and ML for a while, introduced me to LangChain. While we were talking about AI, he mentioned LangChain and said, “It’s like Laravel, but for AI!” Since Laravel is my main framework, that really caught my attention. So, I paused everything I was learning and jumped into LangChain.

Within a week or so, I had a grasp of the basics and started building a small AI app using Python, Streamlit, and LangChain. This app takes a YouTube URL, fetches the transcript, and lets you chat with the video’s content — you can ask the bot about what was discussed, key points, and anything related to the video.

It’s a small win, but for me, it’s a huge one. From getting a “Hello World!” response from my chatbot to building a complete conversational model that can now answer questions properly — that’s a big win in my journey.