Task & Staff Scheduler
The Challenge
We worked with a client in the construction space to build an AI solution that would optimize their construction projects. The client wanted to build an optimization tool that would optimally schedule tasks among their workers to reduce time waste, increase efficiency, and ultimately lower costs.
The Solution
This problem is actually an extension of the Job Shop Scheduling (JSS) problem, which is an NP-Hard problem referred to as a “clearly harder than the Traveling Salesman problem.” This system goes beyond the difficulty of the JSS problem via the introduction of more than 5x the amount of variables we had to optimize over, examples of which include individual staff skillsets, drive time, site visit ordering (traveling salesman), task ordering, resource availability, wage spend, and more.
This vast set of optimization targets actually makes the problem space too difficult for current widely accepted AI solutions such as Neural Networks, which is where GAs come in. Their inference runtime is longer than Neural Networks, but with this they have the capacity to search vastly large parameter spaces to find semi-optimal solutions. We implemented a Genetic Algorithm based solution that was able to consider these multitude of factors effectively to optimally provide scheduling.
The Outcome
This was a particularly challenging problem from a technical aspect, but we were able to develop a solution that not only worked, shaved millions off of our client’s bottom line. These type of results wouldn’t have been possible without leveraging AI and in particular GA, and we rocketed their company and its efficiency forward by leaps and bounds.
