Prakya/Queue-Waiting-Time-Optimizer-V3

Queue Waiting Time Optimizer V3

Queue Waiting Time Optimizer V3 is an AI decision-support system that learns when to open, hold, or close service counters to reduce queue length and waiting time under changing demand.

Built for hackathon impact, this project combines:

• —A custom queue simulator and Gymnasium environment
• —A rule-based baseline policy for reliability
• —A DQN reinforcement learning agent for adaptive control
• —An interactive Gradio dashboard for real-time scenario testing

Why This Matters

Long queues cost revenue, customer trust, and staff efficiency. Most fixed staffing rules are too rigid for real-world demand spikes.

This project introduces adaptive queue control that can react to dynamic arrivals and balance service quality with operational cost.

Problem Statement

Given a live queue system with variable incoming traffic, decide at every step whether to:

• —Close a counter
• —Do nothing
• —Open a counter

Goal:

• —Minimize average waiting time
• —Minimize queue length
• —Avoid unnecessary counter churn and overstaffing

Solution Overview

The system models queue operations as a sequential decision-making problem:

• —State: queue length, waiting time, open counters, incoming rate
• —Action space: {close, hold, open}
• —Reward: weighted combination of waiting time, queue length, number of open counters, and action smoothness

The app supports three policies:

• —Baseline: deterministic rule-based thresholds
• —Random: control sanity check
• —DQN: learned policy (loads model if available, falls back safely otherwise)

Key Features

• —End-to-end RL pipeline from simulation to training and evaluation
• —Scenario-based testing (easy, medium, hard)
• —Seeded reproducibility support
• —Rich visual analytics: queue, waiting time, open counters, cumulative reward
• —Hugging Face Space-ready Gradio UI
• —Safe fallback to baseline when model artifact is unavailable

Tech Stack

• —Python 3.10+
• —Gymnasium
• —Stable-Baselines3 (DQN)
• —PyTorch
• —Gradio
• —Plotly + Pandas

Project Structure

queue-waiting-time-optimizer/
	app.py                        # Gradio app entrypoint
	inference.py                  # Structured inference entrypoint (START/STEP/END)
	configs/base.yaml             # Core reward/scenario config
	scripts/                      # Phase-wise demos, training, eval, plotting
	src/gradio_app.py             # UI and simulation workflow
	src/qwt_optimizer/            # Core package
		agents/rule_based.py
		envs/queue_simulator.py
		envs/queue_gym_env.py
		utils/seeding.py
	tests/                        # Unit tests for simulator/env/reproducibility/agent

Quick Start (Local)

cd queue-waiting-time-optimizer
python -m pip install -r requirements.txt
python -m pip install -e .
python app.py

Open the URL printed in terminal (for example: http://127.0.0.1:7860).

Run Training and Evaluation

cd queue-waiting-time-optimizer

# Train DQN
python scripts/phase5_train_dqn.py --scenario medium --timesteps 20000 --seed 42

# Evaluate baseline vs random vs dqn
python scripts/phase6_evaluate_agents.py --scenario medium --episodes 10 --max-steps 300

# Plot traces and summary charts
python scripts/phase7_plot_results.py

Demo Flow (Hackathon Pitch)

1. 1.Start on medium scenario with baseline policy.
2. 2.Show queue/wait trends and cumulative reward.
3. 3.Switch to DQN policy with saved model path.
4. 4.Compare how policy behavior changes under hard scenario.
5. 5.Highlight reduced congestion and adaptive counter management.

Reproducibility

• —Global seeding is applied for deterministic behavior where possible.
• —Tests cover simulator behavior, Gym environment API, rule-based policy, and reproducibility checks.

Deployment

This repository is structured for Hugging Face Spaces with Gradio runtime.

Space URL:

• —https://huggingface.co/spaces/Prakya/Queue-Waiting-Time-Optimizer-V3

Future Improvements

• —Multi-objective optimization with dynamic reward balancing
• —Traffic forecasting integration for proactive staffing
• —Multi-queue/multi-site coordination
• —Human-in-the-loop controls for operations teams

Team

Queue Waiting Time Optimizer Team

License

For hackathon/demo use. Add your preferred OSS license before production release.