Mintmind pitch video

https://bitcointv.com/w/mTLWUXc9FY1yX8pMP7Rdk8

Please review at your leisure. :)

I followed this tutorial and modified it for Fedimint: https://www.youtube.com/watch?v=CF5buEVrYwo

The repository is here: https://github.com/EthnTuttle/fedimint-chat-bot

I used whisper.ccp (https://github.com/ggerganov/whisper.cpp)) to transcribe several podcast episodes featuring Fedimint and Fedi members.

I consolidated these text files with Markdown documents from the Fedimint.org website.

I then used LangChain (https://www.langchain.com/)) to load this corpus of data into memory before using another LangChain library to prepare embeddings (OpenAI Embeddings) from the files, which were then uploaded into Pinecone (https://www.pinecone.io/)..) This vector database can now be queried for contextual results only from our data and can be updated with newer relevant data over time. These results are then fed into a "chain", along with an LLM (we used ChatGPT for this, but a locally hosted one could be used instead). Then a call is made to the LLM using the vectorize context to ensure the LLM does not "hallucinate" aka make things up. This also means the LLM can have more current data than when it was trained.

Motivation:
I started contributing to Fedimint earlier this year and have been looking for ways to best contribute. I am much better at running code than writing code at this point so have focused on enabling users by contributing to UI and other ease of use features for Fedimint. After a discussion about this hackathon, I realized that my recent exploration of LangChain during fiat mining gave me a way to contribute to the hackathon and Fedimint by making a Q&A bot. I could use LangChain to surface information about the complex topic of Fedimint. I later learned about wrapping web functions in LN payments which was a last minute feature add that now makes Mint Mind sustainable over a longer term.

I intend to keep Mint Mind running as long as it remains cost neutral. I see it as a way to stack more sats by spending fiat for servers/API and receiving sats for code execution. After receiving some feedback, I made the sats amount selectable and lowered the minimum amount so more people can benefit from Mint Mind.