Inspiration
There is much we can learn from the thousands of research papers and firsthand accounts of "spiritual" phenomena to understand the reality we live in, but all this information is overwhelming and not currently accessible to the average person. Spiritual Data's mission is to change that through leveraging Generative AI and embeddings.
What it does
For this prototype of our upcoming MVP we decided to build a chat bot that answers the user's questions using multiple datasets: research papers, spiritual experiences, and hypotheses.
How we built it
We use OpenAI to parse out the relevant parts of the user's input to the chatbot to create embeddings. We also have a chatbot UI with Clerk for user management and Netlify and AWS for deployment.
Challenges we ran into
Associating metadata with each segment of text. Adapting a Langchain retriever to do multiple searches in Pinecone.
Accomplishments that we're proud of
- Using embeddings of segments of text to provide the chatbot and user with the most relevant text.
- We're open-source!
What we learned
Pinecone is nice as a cloud solution with a reliable central store of embeddings, making it easy to collaborate. We can save on costs by using multiple namespaces in one index, and also filter by metadata.
What's next for Spiritual Data Chatbot
Finishing the MVP and using embedding search to match hypotheses to research and experiences so we can calculate reliable probabilities for scientific questions! Sending a pull request to LangChain for our multisearch retriever and conversational chain.
Built With
- amazon-ec2
- amazon-web-services
- langchain
- notion
- openai
- pinecone
Log in or sign up for Devpost to join the conversation.