Vector Store with ChromaDB

Continuing our discussion about Retrieval Augmented Generation (RAG), this week we will incorporate ChromaDB as the storage and query mechanism for the embeddings. Our initial SpaceApps submission held all of the embeddings in memory and used a cosine similarity metric for queries. We will go through the ChromaDB API and design decisions required to replace the custom approach with a more robust solution.

The initial work for converting this is posted in the SpaceApps repo linked below if you want to check it out early. Also – I double-checked our availability at HudsonAlpha, so we are set for an in-person meetup if you can make it.

Series:

Over the next several sessions, we will be diving deeper into separate components needed for RAG – hopefully resulting in a chat-based Q&A service for the NASA Technical Report Server.  We were introduced to this data during our submission for the 2022 NASA SpaceApps Challenge – where we placed 2nd. Our submission was a semantic search based on the abstracts for the NSTR dataset of 10,000 papers.

I hope to have the video from our last session posted today. You can check for updates at https://hsv.ai/videos

Links:

• Chroma DB – https://github.com/chroma-core/chroma
• RAG Paper – https://arxiv.org/pdf/2005.11401.pdf
• Huntsville AI 2022 SpaceApps Submission – https://github.com/HSV-AI/spaceapps2022

Details:

• Date – 03/20/2024
• Time – 6-7pm
• Location – HudsonAlpha
• Address –  601 Genome Way Northwest, Huntsville, AL 35806
• Zoom –https://us02web.zoom.us/j/81139026249?pwd=bTR1VUoxRGMrN3I1VnN1NVRDNkdIZz09