Challenges

This personal project started as a playful experience, but it was grounded in a real observation:

While there are countless AI chatbots available, very few aim to be genuinely lighthearted.

I wanted to explore what it might look like if a chatbot was designed primarily to be funny - while still feeling responsive and engaging.

Solution

I designed a personality that does not just drop random jokes, but responds in context and stays consistent throughout the conversation. I wanted to make humour feel intentional and not just chaotic AI randmness.

That required appropriate prompt design and an understanding how tone can shift subtly depending on the topic or user input. Running this on a free LLaMA model added a few technical constraints around content length. Thus, I implemented ways to keep conversation history flowing, manage prompt size, and monitor usage to avoid performance dips.

My Role

As a personal project created for fun, I designed and built the chatbot end-to-end. This gave me space to use my full range of my skills, from shaping the user experience to solving technical challenges under real-world constraints.

This project reflects what I enjoy the most - hands-on experimentation with emerging tools, guided by curiosity and a strong sense of how product, design and technology intersect.

Results

While PunGPT won’t win any awards for productivity, it delivers a steady stream of puns and dad jokes and the occasional eye-roll - just as intended.