Vertex AI RAG RFP Processor

Overview

This cloud function automatically processes Request for Proposal (RFP) Excel sheets using Google's Vertex AI and Discovery Engine. It transforms requirements into questions, generates responses using RAG (Retrieval-Augmented Generation), and provides referenced answers from your knowledge base.

Features

• 🤖 Automated RFP response generation
• 📝 Answers questions based on the RFP requirements
• 🔍 RAG-based answer generation using Discovery Engine
• 📚 Provides source references for each answer
• 📊 Maintains Excel formatting with proper styling
• ♻️ Automatic retry mechanism for API failures
• 🔄 Processes multiple sheets within a workbook

Prerequisites

• Google Cloud Project with enabled APIs:
  • Vertex AI API
  • Discovery Engine API
  • Cloud Storage API
• Python 3.7+
• Required Python packages:

  functions-framework
  google-cloud-storage
  pandas
  vertexai
  openpyxl
  requests
  google-auth

Setup

1. Clone this repository
2. Set up your Google Cloud project:

   export PROJECT_ID="your-project-id"
   gcloud config set project $PROJECT_ID

3. Enable required APIs:

   gcloud services enable \
       aiplatform.googleapis.com \
       discoveryengine.googleapis.com \
       storage.googleapis.com

4. Configure Discovery Engine with your knowledge base
5. Update the project configurations in the code:
   • Update project_id in initialize_vertex_ai()
   • Update Discovery Engine URL in call_discovery_engine()

Usage

1. Upload your RFP Excel file to the configured Google Cloud Storage bucket
2. The cloud function automatically triggers and processes the file
3. A new file with prefix processed_ will be created in the same bucket
4. Download the processed file containing:
   • Original requirements
   • Generated questions
   • RAG-based responses
   • Reference sources

Excel File Format

The function supports Excel files with the following column names:

• Bank Requirement or requirements - Contains the RFP requirements
• Additional columns will be preserved, and new columns will be added:
  • Questions - Generated Yes/No questions
  • Vendor Response - RAG-generated answers
  • References - Source documents references

Error Handling

• Implements exponential backoff retry mechanism
• Handles API failures gracefully
• Skips already processed files
• Validates sheet structure before processing

Limitations

• Processing time depends on the number of requirements
• 20-second delay between RAG queries to respect API limits
• Maximum 3 reference sources per response
• Answers are limited to 100-200 words

Authors

Tejan Gavandi

Acknowledgments

• Google Vertex AI
• Google Discovery Engine
• Cloud Functions Framework