Hotel Search Engine

Portfolio-scale hotel search and ranking system inspired by Booking.com.

Status

• UI: FastAPI query parser with mock ranking.
• Query understanding: preprocessing, budget classification (ONNX), flexibility detection.
• Data and ML pipelines: placeholders.

Quick Start

make dev       # start Duckling (Docker) in background + FastAPI app

Then open: http://127.0.0.1:8080

Make Commands

Command	Description
make run	Start the FastAPI app on port 8080
make duckling	Start the Duckling date-parser container (port 8000)
make dev	Start Duckling in background, then launch the app
make train	Train the budget classifier, log to MLflow, export to ONNX
make mlflow-ui	Open the MLflow tracking UI at http://localhost:5000
make install	pip install -r requirements.txt
make install-dev	Install dev extras: pytest, ruff, jupyter
make venv	Create the hotelenv virtual environment
make notebook	Open Jupyter Lab in notebooks/
make test	Run the test suite
make test-v	Run tests with verbose output
make lint	Lint source with ruff
make format	Auto-format source with ruff
make data-info	Print a summary of the raw CSV dataset
make model-info	List saved ONNX models
make clean	Remove __pycache__ and .pyc files
make clean-all	Remove cache and delete hotelenv entirely

Run make help to see this list at any time.

MLflow — Experiment Tracking & Model Registry

Every make train run logs to a local SQLite database (mlflow.db) and registers the model.

make train       # train + log metrics + register model
make mlflow-ui   # open http://localhost:5000

What gets tracked per run:

Category	Details
Parameters	TF-IDF settings, MLP architecture, train/test split
Metrics	Accuracy, macro precision/recall/F1, per-class breakdown
Artifacts	budget_classifier.onnx
Model Registry	budget-classifier — version increments on every run

mlflow.db and mlruns/ are in .gitignore — not committed to version control.

Manual Setup (without make)

# 1. Create venv and install dependencies
python -m venv hotelenv
hotelenv/Scripts/pip install -r requirements.txt

# 2. Start the Duckling date parser
docker-compose up -d duckling

# 3. Train the budget classifier (first time only)
hotelenv/Scripts/python -m src.query_understanding.budget_classifier

# 4. Start the app
hotelenv/Scripts/python -m uvicorn src.app.streamlit_app:app --reload --port 8080

Requirements

• Python 3.12+
• Docker (for Duckling)
• make (Git Bash / WSL on Windows)