Private AI For GCP

Created by team State Change Labs on July 09, 2023

Medium sized companies that are GCP customers have a problem. They hear about the great power of OpenAI and want to take advantage to leverage the technology for their businesses. But at the same time, they have significant proprietary data that they can't hand to just anyone, and they are concerned about the use of this information in the hands of startups they've never heard from. Howe can we solve this problem? The new releases in Vertex mean an opportunity to re-connect the dots of modern generative AI technology with the information they have today without ever leaving Google. Our implementation connects audio import of meeting recordings using Google Cloud Run and Google Speech-To-Text using the latest-generation Chirp model. We then load that data into Google CLoud Storage and apply it to a Vertex matching engine via the Gecko embeddings model. The data is queryable using a langchain-enabled Chainlit instance again running on Cloud Run which leverages both the latest Palm 2 chat interface and vector search from Matching Engine. Now instead of using Pinecone, Ada and GPT4, one can use an all-google approach for compliance and safety.

Category tags:

"Great concept! I would have liked to see a small working demo, on dummy data samples or something like that."


Chinmay Jog

Machine Learning Engineer