Flight-Insight-AI

Introduction

Flight Insight AI is a chatbot system designed to provide intelligent insights and analysis of airline reviews. Using natural language processing and machine learning techniques, Flight Insight AI offers users a seamless interface to query and understand airline experiences based on real customer reviews. Try it Here!

System Architecture

• Frontend: A React application providing a responsive and interactive user interface.
• Backend: A FastAPI server that handles real-time communication and orchestrates the AI components.
• Vector Database: Pinecone is used for efficient similarity search of review embeddings.
• Language Model: Google's Gemini Flash for natural language understanding and generation.
• Data Processing: Several scripts to get and clean the data.

Sequence Diagram

1. User Input: The user types the query into the Frontend interface.

• Query Transmission: The Frontend sends this query to the FastAPI Backend via a WebSocket connection.

2. Airline Extraction:

• The Gemini Flash model for airline extraction identifies the airline in question.

3. Query Embedding:

• The processed query is sent to the Query Embedder. The Query Embedder converts the text query into a vector embedding, similar to how the reviews were embedded.

4. Relevant Review Retrieval:

• The Retriever takes this query embedding and searches the Pinecone Vector DB for similar review embeddings. It retrieves the TOP_K relevant reviews similar to the user query.

5. Prompt Generation:

• The Prompt Generator receives:

  • The original query
  • The relevant reviews from the Retriever

• It combines this information into a structured prompt for the language model.

6. Response Generation:

• The generated prompt is sent to the Gemini Flash model for response generation. This model processes the prompt and generates a comprehensive response based on the relevant reviews.

7. Response Delivery:

• The generated response is sent back to the Frontend via the WebSocket connection. The Frontend streams this response to the user.

Features

• Output Streaming: Utilizes WebSocket for streaming AI responses, providing a dynamic and engaging user experience.
• Intelligent Review Analysis: Leverages a RAG (Retrieval-Augmented Generation) system to provide insights based on relevant customer reviews.
• Dynamic Airline Information: Automatically detects airline mentioned and responds based on that.
• Vector Search: Pinecone DB is used for efficient similarity search of review embeddings, ensuring fast and relevant retrievals.
• LLM: Integrates Google's Gemini Flash for high-quality natural language processing and generation.

Technology Stack

• Frontend: React, Tailwind CSS
• Backend: FastAPI
• Vector Database: Pinecone
• Language Model: Google Gemini Flash
• Embedding Model: SentenceTransformer
• Deployment: Vercel, Render (https://flight-insight-ai.vercel.app/)