RupeeTravel - Conversational Query Interface

RupeeTravel is a backend API for conversational flight search, powered by an LLM and a local flight database. Users can interact using natural language queries to find flights, get prices, and specify trip details.

Features

• Conversational Flight Search: Accepts free-form queries (e.g., "Find me the cheapest flights from Delhi to Hanoi in December").
• LLM-Powered Intent Extraction: Uses Qwen3:1.7b via Ollama for robust query understanding.
• Guardrails: Python logic corrects common LLM extraction mistakes for reliability.
• SQLite Database: Stores flight data locally for fast queries.
• FastAPI Backend: RESTful API endpoints for integration.
• Automatic Database Population: Loads flight data from flight-price.json on startup.

Frontend

A companion frontend for RupeeTravel is available at:
https://github.com/harsh-vardhhan/flight-search-frontend

Tech Stack

• Language: Python 3.11+
• Framework: FastAPI
• LLM: Qwen3:1.7b via Ollama
• Database: SQLite (flight.db)
• ORM: SQLAlchemy
• Prompt Engineering: LangChain
• Environment Variables: Managed via .env

API Endpoints

• POST /transcript
  Accepts a user query and returns matching flight options or clarification questions.

• GET /
  Health check endpoint.

File Structure

• main.py — FastAPI app and core logic
• llm_logic.py — LLM prompt and intent extraction chain
• crud.py — Database query functions
• models.py — SQLAlchemy models
• schemas.py — Pydantic schemas
• database.py — DB setup
• flight-price.json — Source flight data

How It Works

1. Startup:

   • Database tables are created.
   • Flight data is loaded from flight-price.json.

2. Query Flow:

   • User sends a query to /transcript.
   • LLM extracts intent (origin, destination, dates, trip type, etc.).
   • Guardrails fix missing or ambiguous info.
   • Database is queried for matching flights.
   • Results or clarifications are returned.

Setup

1. Install dependencies:

   pip install -r requirements.txt

2. Start Ollama with Qwen3:1.7b:

   ollama pull qwen3:1.7b
   ollama serve

3. Run the API:

   uvicorn main:app --reload

Environment Variables

Configure your .env file for any required secrets or LLM settings.