The problem Greeting Master solves

Limited options for creating personalized, multimedia greetings on WhatsApp.

Existing tools often require additional apps and technical know-how, making them inaccessible to many users, especially the older generations

Challenges we ran into

🚧 Token Length Limitations

One of the issues we faced was related to the token length limitations of the GPT models. We had to carefully manage our input so that we did not exceed the maximum token limit. Through multiple iterations and testing, we devised a method to split the input text into chunks that could be processed separately, without losing coherence.

🗣️ Autonomous Chatbot with GPT-4

Creating an autonomous chatbot using GPT-4 was a challenge due to the model's complexity and resource requirements. We made multiple iterations on the chatbot's implementation, focusing on prompt engineering, to ensure the bot's responses were appropriate and coherent. JSONifying the response from GPT-3.5/4 was another task that required careful handling to ensure the data was correctly parsed and used.

⏱️ Latency Issues

We also faced latency issues with GPT-4 due to its high computational requirements. To mitigate this, we performed multiple iterations on the chatbot, focusing on reducing the latency. We streamlined the number of messages exchanged in each workflow to a minimal amount, thus improving response times.

🎥 Video Generation

For video generation, we tried multiple services including D-ID and SadTalker. While SadTalker provided valuable learnings and was used to generate a number of videos, we found D-ID to be faster for our use case and thus decided to proceed with it.