"Queue"-Learning: An Adaptive Queueing Workflow

Inspiration

Queueing in modern stores is incredibly difficult for human representatives to handle manually. Naively, a human representative would have to keep track of the order in which customers enter the store. However, this is nowhere near an optimal arrangement: there is a benefit, for example, to allowing short 2-5 minute visits - for quick fixes, like swapping a SIM card - ahead of longer, more involved customer-employee relationships such as in sales. We have developed an integrated system to allow customers to easily register themselves into a queue as well as a novel queuing algorithm inspired by reinforcement learning techniques to reduce mean/median wait times.

What it does

"Queue"-Learning uses simulated annealing to compute a near-optimal schedule for a cost functional chosen to optimize costomer experience.

How I built it

We used python for the backend and react for the frontend.

Accomplishments that I'm proud of

Our algorithm outperforms the mean results of the naive algorithm by nearly 2x in high-density scenarios, and likely performs even better when dealing with ahead-of-time reservations.

Results

In a randomized simulation with heavy traffic, our algorithm averaged a median customer wait time of 17 minutes while the "default" naive algorithm, which might be carried out by a representative in a store, took an average of 27 minutes.

Built With

• python
• react