EchoMind

Inspiration

How do we train a chatbot to achieve a specific goal? Chatbots are great at generating specific output but lack consistency, and this is the idea behind our project.

What it does

It stores prompts for ChatGPT in a database and grows a "memory"/"identity" based on prompts that directly impact different areas associated with its "e-personality." It's packed with the personality of a Martian called Marian Martianu, and this is where things get interesting. Marian distinguishes between core memories—areas that it focuses on—and other prompts to keep its DISTINGUISHED identity and SPARKLING personality. Marian is a unique individual. As the creators of this universal being, we're also monitoring his "brain", keeping things in check, ready to unplug Marian if needed. - Eat your heart out, Skynet.

How we built it

We leveraged ChatGPT prompt engineering to build a persona and then enhanced that persona with custom context building. This implies seeding memories into the persona and saving each new interaction as a memory. Each memory has associated metadata, and the intersection of available metadata with input metadata results in "memory retrieval."

Challenges we ran into

Making this project stand out and adding extra panache

Accomplishments that we're proud of

Visual identity of the project Using a simple idea but packaging it in a fun way Doing a lot of prompting and only a little coding, including generating visuals