RAG-powered survey analysis tool

Demo

Table of Contents

• Overview
• Tech Stack
• RAG agent workflow
• Sample of queries
• Services topology diagram
• Frontend
  • Frontend dependencies
  • Frontend Severless settings recommendations
• Backend
  • Services
    • Index
    • Chat
  • Backend dependencies
  • Backend Severless settings recommendations

Overview

This project involves developing a web application that enables users to explore, analyze, and compare two market research reports through AI-powered natural language interactions. The system architecture consists of a Python backend built with FastAPI, which serves as the bridge between the frontend interface and a Retrieval-Augmented Generation (RAG) system. The RAG system leverages advanced techniques such as vector embeddings, semantic search, and prompt engineering to retrieve relevant information from the reports and generate insightful, contextual responses. The application must ensure all AI-generated results are traceable, allowing users to double-click or access the original source data for verification and deeper analysis.

This software demonstrates Retrieval Augmented Generation (RAG) for in-context learning using specific PDF documents:

• 2023 Analyst & Investor Meeting.
• 2024 Proxy Statement.

Tech Stack

• LLMops: LanghGraph.
• Frontend: ReacJS, Material.
• Backend: FastAPI.
• Database: Chroma.
• Infrastructure: Google Cloud Platform (GCP), Terraform, Makefile, Docker, Docker Compose, CloudRun for serverless architecture, Firebase.
• Quality: ESlint, pylint, pytest, pycov.

RAG agent workflow

Sample of queries

• Please tell me about the recognitions and achievements in 2023.
• Who is Mr. Murti?
• What is Cash from operations for?
• How many years of relationship does ConocoPhillips have with Qatar?

Services topology diagram

Frontend

Frontend dependencies

• "@emotion/cache": "^11.9.3"
• "@emotion/react": "^11.9.3"
• "@emotion/styled": "^11.9.3"
• "@mui/icons-material": "^5.8.4"
• "@mui/lab": "^5.0.0-alpha.88"
• "@mui/material": "^5.8.6"
• "@mui/system": "^5.8.6"
• "@mui/utils": "^5.8.6"
• "@tabler/icons": "^1.72.0"
• "axios": "^1.3.6"
• "crypto-js": "^4.0.0"
• "formik": "^2.2.9"
• "framer-motion": "^6.3.16"
• "i18next": "^22.4.13"
• "node-polyfill-webpack-plugin": "^2.0.1"
• "prop-types": "^15.8.1"
• "react": "^18.2.0"
• "react-detect-offline": "^2.4.5"
• "react-device-detect": "^2.2.2"
• "react-dom": "^18.2.0"
• "react-gtm-module": "^2.0.11"
• "react-i18next": "^12.2.0"
• "react-loading-skeleton": "^3.2.1"
• "react-perfect-scrollbar": "^1.5.8"
• "react-redux": "^8.0.2"
• "react-router": "6.3.0"
• "react-router-dom": "6.3.0"
• "react-scripts": "^5.0.1"
• "react-tooltip": "^5.28.0"
• "redux": "^4.2.0"
• "yup": "^0.32.11"

Frontend severless settings recommendations

• Memory (Memory to allocate to each container instance): 128Mb
• CPU (Number of vCPUs allocated to each container instance): 0.1
• Request timeout (Time within which a response must be returned): 300 seconds
• Maximum requests per container (The maximum number of concurrent requests that can reach each container instance): 1
• Minimum number of instances (Autoscaling): 1
• Maximum number of instances (Autoscaling): 5

Backend

Services

Index

Category: Endpoint
Relative URL: /
Description: App description.

Chat

Category: Endpoint
Relative URL: /chat
Description: In-context ChatGPT-based chat.

Backend dependencies

• python3.11
• python3-pip
• python3.11-distutils
• python3.11-dev
• langchain==0.3.15
• onnxruntime==1.20.1
• chromadb==0.6.3
• langchain-chroma==0.2.1
• langchain_community==0.3.15
• langgraph==0.2.67
• pypdf==5.1.0
• tiktoken==0.8.0
• gpt4all==2.8.2
• langchain-openai==0.3.5
• pysqlite3-binary
• fastapi[standard]
• pylint == 3.3.4
• pytest == 8.3.4
• pytest-cov == 6.0.0

Backend severless settings recommendations

• Memory (Memory to allocate to each container instance): 512Mb
• CPU (Number of vCPUs allocated to each container instance): 1
• Request timeout (Time within which a response must be returned): 300 seconds
• Maximum requests per container (The maximum number of concurrent requests that can reach each container instance): 1
• Minimum number of instances (Autoscaling): 1
• Maximum number of instances (Autoscaling): 5

🔼 Back to top