STORY
ChatBTC - pitch
AUTHOR
Joined 2023.07.12
PROJECT
DATE
VOTES
sats
COMMENTS

ChatBTC - pitch

🤖 ChatBTC

The journey to learning to build on bitcoin is one that we know feels lonely, frustrating, and sometimes hopeless. 

It’s like there’s no one to really talk with to flesh things out, to dive deeper. Someone that really gets it in a high-quality way. 

Until now. 

We’ve built ChatBTC on the highest signal sources in the Bitcoin ecosystem, including: Bitcoin Transcripts, Bitcoin and Lightning Dev mailing lists, Bitcoin Optech, among others. 

Not only does the chatbot provide our learners with high-signal outputs… but it references the relevant places sources where this information was gleaned.

And to revv the engine just a bit further - we also build AI conversation bots from some of the most prolific authors in bitcoin like Matt Corallo, Andrew Chow, and Greg Maxwell. (They all gave their consent!).

One last thing…

To protect from DDOSing and to encourage a circular economy we used ⚡webLN⚡ to paywall ChatBTC responses after some freebies.

đź‘€ See for Yourself

https://youtu.be/sFgGqmWezZs

🎢 Experience for Yourself

Check out our live product below! 

⚒️ How We Built This

  • We built using light scrum methodology cycles of “plan, build, review, adjust”

  • All our code is for the public to play and fork on GitHub

  • We ongoingly solicited feedback from users to ensure we’re meeting and exceeding their needs (we’re building for them after all!)

  • We used surveys and moderated usability tests to get feedback about our products

đź’˝ Technical Stuff

  • nextJS app that makes queries to an ElasticSearch cluster and uses the sources to create an answer by using OpenAI gpt-3.5-turbo

  • We use voltage to host our LND node and have alby integration for the paywall.

  • We log questions, answers and ratings on supabase. (We don’t collect any other user info.)

⏩ And Then What?

Now

  • Chat with bots that respond based on the Lightning paywall

Next

  • Adding more authors and improving results

Later

  • Using Q&A to train better models

đź«‚ Team