Linkd Social | Devpost

Inspiration

Large-language models have taken the world by storm. From the incredibly versatile and useful ChatGPT, to more niche and experimental use cases like the popular 24/7 streaming show watchmeforever on twitch, there is hardly an area of the technology or entertainment industry that hasn't been touched by the incredible power of these models. What if people could leverage these models to become an extension of themselves, and connect with their fans and friends, even when they cannot be present? This is the inspiration of Linkd: Exploring human connection.

What it does

In the demo, there are 24 fictitious personas to talk with. A revenue sharing contract distributes tokens every 24 hours using Chainlink Automation and the Chainlink oracle fetch API.

How we built it

The project is built in Typescript to run in a NodeJS environment. It uses the GPT Completions API from OpenAI in conjuction with custom prompts and dialogue samples to mimic a cloned persona. Smart contracts and token were built in Solidity with leveraging Chainlink in mind.

Challenges we ran into

A core challenge is time constraints, as a solo entrepreneur, I also work a day job, and have a demanding schedule. On the technical side, there were challenges regarding how to structure the application in a modular fashion that is suitable to run as a self-contained Telegram bot. The challenge was addressed by using Singleton instances of the data store, and electing to run components as modules instead of separate node scripts.

Accomplishments that we're proud of

I am proud of the intricacy and entertainment value of the platform that I've built and I am looking forward to seeing it grow and expand!

What we learned

I learned a lot about how to parse and process the outputs from LLMs effectively, as well as how to craft engaging and personable prompts. I also discovered the utility chainlink can bring to applications that are not typically thought of as being strictly web3.