harinBot

harinBot repository:

embedchain contribution:

Based on prompt engineering, harinBot leverages the embedchain framework and my entire Facebook, WhatsApp and Discord message history to make a chatbot mimic my conversation styles.

This included solving bugs with the embedchain framework and submitting an approved PR into a project with over 3.5k stars.

In short, my conversations are embedded into tokens, stored on a database. When a user talks to the chatbot, their sentence is embedded as well and queried for similar vectors, and those sentences are then used as context for the GPT inference model provided by OpenAI.

This allowed me to explore the power of prompt engineering and the latest frameworks in NLP. The end result provided surprisingly similar responses to my own, which are redacted here for privacy. However, limitations could be detected as GPT's extremely polite writing style was very much different from everyday conversations with friends. I'm hoping to incorporate my own models in the future and help make more improvements on the rapidly growing embedchain project.