search

SartajBhuvaji / Resonate

Data Science Capstone Project based on RAG LLMs. The project aims to improve meetings by providing an interface to recollect information from audio/video meetings.
https://huggingface.co/spaces/sartajbhuvaji/Resonate
MIT License
6 stars 1 forks source link

Added initial notebooks : Pinecone DB #6

Closed SartajBhuvaji closed 7 months ago

SartajBhuvaji commented 8 months ago

Added notebooks to play around with Pinecone

  1. pinecone_trial_1.ipynb getting to know pinecone and understanding the basic functions

  2. pinecone_demo_transcript.ipynb implementing data store and querying using a demo transcript

PrachiteeChouhan31 commented 8 months ago

The scores are good. but I was thinking can we join adjacent rows with same speaker together and then get embedding.

jay-singhvi commented 8 months ago

This could increase the content we send to llm for processing but it also means llm will get more context about the topic

SartajBhuvaji commented 8 months ago

@PrachiteeChouhan31 @jay-singhvi Yes, that's a great idea!

Let us decide the context window size and then in v2 we can send out a second query to Pinecone to fetch data in that window. This would help us get the relevant context of the conversation too.

madhuroopa commented 8 months ago

@SartajBhuvaji are you talking about the + secs and - secs content to be retrieved when queried ?

SartajBhuvaji commented 8 months ago

@madhuroopa Yes. +/- window size data.