✈️ Airline Policy and Flight Query Chatbot: Program Overview

This program implements a chatbot system that combines Retrieval-Augmented Generation (RAG) and API-based flight search to answer user queries related to airline policies and flight availability. It processes user input to classify and respond appropriately, leveraging LLMs, vector databases, and external APIs.

---

🚀 Key Features

1. Query Classification:

   • Classifies user queries into three categories:
     • RAG-Related: Queries about travel policies (e.g., luggage rules, traveling with pets, etc.).
     • API-Related: Questions requiring live flight information (e.g., cheapest flights, flight performance).
     • General: Queries outside the chatbot's scope.

2. Policy Retrieval:

   • Uses LLamaIndex with ChromaDB as the vector store for retrieving policy documents.
   • Handles both specific airline queries and general ones using metadata filtering.

3. Flight Search:

   • Supports various flight queries via APIs, including:
     • Flight availability between two cities.
     • Cheapest flight offers.
     • Performance of specific airports.

4. Adaptive Query Engine:

   • Dynamically configures the query engine based on the classification of the input (e.g., specific airline queries use metadata filtering).

5. Multimodal Input Handling:

   • Accepts free text from the user and extracts structured information (e.g., origin, destination, and date) using LLMs.

---

Present Flow:

1. Main
2. Create RAG Index
   • Define the index, vector store, and retriever while providing metadata.
   • Current metadata includes:
     • Bag, pets, and sports equipment policies of 3 airlines:
       • American Airlines
       • Delta Airlines
       • United Airlines
3. Prompt Handling
   • Continuously asks for user input until the user says exit, quit, or bye (currently hard-coded, can be enhanced with LLM).
4. Query Classification
   • RAG Query:
     1. Call create_prompt.
     2. Execute classify_rag_query to determine the relevant metadata.
     3. Build the query using the RAG output.
     4. Return the result.
   • API-Related Query:
     1. Classify the type of API call required:
        • Call FlightAvailBtwCities for flights between two cities on a specific date.
        • Call FlightInspirationSearch for the cheapest flights from a location.
        • Call FlightOffersSearch for the cheapest flights between two cities.
   • Other Queries:
     • Return a response indicating the chatbot cannot answer the question.

---

Questions Answered:

1. FlightAvailBtwCities

Handles Questions:

• Flights between two cities on a specific day.
• Number of seats available on a flight between two cities.
• Filtering based on:
  • Fastest flights.
  • Flights below a specific duration.
  • Non-stop flights.
  • Time of day (may be ambiguous).
  • Avoiding particular layovers (e.g., country-based).
  • Avoiding or preferring specific carriers.

Issues:

• number_of_stops always returns 0. Needs adjustment to reflect duration.
• Multiple flights are represented as a second instance of a segment.

---

2. FlightInspirationSearch

Handles Questions:

• Cheapest flights from a specific location within a date range.
• Flights based on a budget (e.g., "My budget is $10,000. What are the cheapest flights?").
• Calls AirlineCodeLookup for comprehensive answers.

Notes:

• ViewBy = Date fetches more options when explicitly mentioned, even though it's a default parameter.

Follow-Up Questions:

• "What are the rates?"

Limitations:

• Location filtering is not available in the input.
• Does not answer "Cheapest flight between two locations" (use FlightOffersSearch for this).

---

3. FlightOffersSearch

Handles Questions:

• Cheapest flights between two cities (sorted by price).
• Filters for:
  • Airlines (include or exclude).
  • Class (ECONOMY, PREMIUM_ECONOMY, BUSINESS, FIRST).
  • Non-stop flights (not working currently).
  • Maximum price per traveler.

Notes:

• Only single-date queries are allowed; multi-query is required for multiple dates.
• Provide just the cheapest option or ask follow-up questions for clarity if input is minimal.

Filters in Output:

• Departure and arrival times.

---

Summary of Issues:

• number_of_stops in FlightAvailBtwCities is returning 0 universally.
• Non-stop filter in FlightOffersSearch is currently non-functional.
• Location filtering in FlightInspirationSearch is absent.
• Multi-date queries require multiple function calls.

---