GCP RAG (Retrieval Augmentation Generation) Vectorstore workshop - PGVector

Final Product: Chat with your Data (Scrape public websites/ your PDFs) GenAI project

This repository is designed for a workshop teach the concept of in context learning and working with vectirstires (pgvector) when building GenAI applications with LangChain.

Tech Stack

Server Side: LangChain 🦜🔗

Vectorstore: Postgres PGVector

Embeddings: GCP VertexAI

LLM: PaLM 2

Environment Variables

To run this project, you will need to add the following environment variables to your .env file

DB_URL APPLICATION_DEFAULT_CREDENTIALS

Run Locally

Clone the project

  git clone https://github.com/emarco177/gcp-llm-workshop.git

Go to the project directory

  cd gcp-llm-workshop

Install dependencies

  pipenv install

Start the Streamlit server

  python3 run rag/ingestion.py
  python3 run rag/retrieval_aguentation.py

NOTE: When running locally make sure GOOGLE_APPLICATION_CREDENTIALS is set to a service account with permissions to use VertexAI

GCP SETUP

Please replace $PROJECT_ID with your actual Google Cloud project ID.

To deploy manually:

1. Make sure you enable GCP APIs:

gcloud services enable vertexai.googleapis.com

2. Create a service account vertex-ai-consumer with the following roles:

gcloud iam service-accounts create vertex-ai-consumer \
    --display-name="Vertex AI Consumer"
    

gcloud projects add-iam-policy-binding PROJECT_ID \
    --member="serviceAccount:vertex-ai-consumer@PROJECT_ID.iam.gserviceaccount.com" \
    --role="roles/serviceusage.serviceUsageConsumer"

gcloud projects add-iam-policy-binding PROJECT_ID \
    --member="serviceAccount:vertex-ai-consumer@PROJECT_ID.iam.gserviceaccount.com" \
    --role="roles/ml.admin"

gcloud projects add-iam-policy-binding PROJECT_ID \
    --member="serviceAccount:vertex-ai-consumer@PROJECT_ID.iam.gserviceaccount.com" \
    --role="roles/vertexai.admin"

🚀 About Me

Eden Marco, Customer Engineer @ Google Cloud, Tel Aviv🇮🇱