How can we better advocate for our discipline? How do we provide practical hands-on experience to students and early-career professionals? How can we apply industry expertise to today’s social challenges? How can we collaboratively solve problems in an engaging way that has greater potential for real-world impact?
While these questions can apply to just about any industry, they’ve been top of mind in operations research and decision science. One way that often comes up is through hosting collaborative hands-on events, which can go by many names: hackathons, code fests, innovation challenges. As any organizer of such events knows, you not only need good content (problems, datasets, etc.), planning, and community engagement, you also need a platform that provides the glue for bringing it all together in person, virtually, or both. I’ve seen this unfold for events such as NASA’s International Space Apps Challenge, and it still rings true for things like Kaggle and citizen-driven initiatives like Zooniverse.
Every so often a question pops up about “Are there OR challenges similar to Kaggle” or “Are there any contests/hackathons for OR?” We hear it regularly from folks in the community: practitioners, partners, researchers, professors, and students. While there is an abundance of challenges ripe for decision science solutions, people who are interested in participating and planning, there aren’t robust platform options available to leverage and adapt to make the innovation magic happen.
In this post, I’ll walk through one way to approach using Nextmv as a platform for hosting hackathon/challenge events appropriate for a variety of audiences.
Create dedicated spaces for teams
Event organizers can spin up dedicated teams within Nextmv and invite participants and collaborators to the appropriate team. For example, the challenge may have a total of 5 teams (Team A, Team B, etc.) with a set number of people per team.
Teams can be created ahead of time with the ability to invite more, including those who might serve as event support or judges of a team’s final deliverable.
Populate spaces with starter code, inputs, datasets
So everyone starts from the same place, team workspaces can be pre-populated with starter model code, input data, and datasets (or sets of input files). For example, you might have vehicle routing model code with input sets for food delivery in the Netherlands that represent deliveries for a given month, varying vehicle fleet sizes, and even subregions within the country.
This can be done for any use case. Just bring your model code (housed in a GitHub repo or notebook) and push it to the appropriate team space in Nextmv. While we provide starter models in our open source community repo, Nextmv also has a number of integrations with open source projects such as HiGHS, VROOM, Pyomo, CVXPY, and Nextplot, which can provide a lower barrier to entry for challenge organizers than immediately starting with a commercial offering we integrate with.
Pre-populating team spaces with relevant inputs and datasets in CSV or JSON format is straightforward. Input sets can be uploaded from another source or derived from previous model runs in Nextmv.
Run models and experiments on turnkey infra
As the teams collaborate on solution development, they can customize the model code and push it to Nextmv's infrastructure for fast deployment and experimentation. The model appears in the Nextmv UI where other teammates can run the model, manage model versions, view run history (did runs succeed/fail?), perform experiments to compare iterations, and view experiment results.
The Nextmv platform automatically creates a suite of API endpoints for running custom decision models. (A nice feature that simplifies how OR practitioners interface with software teams in industry, by the way.)
Everyone moves together as models get developed on managed infrastructure that just works, you don’t have to worry about provisioning resources, and doesn’t cause computer fans to kick into high gear.
Run experiments to benchmark and compare results
Nextmv’s experimentation tools and results visualization provide a useful evaluation experience for hackathon teams as they work on their challenge project and help organizers for final judging. For example, teams can run tests that simulate “what if” questions with scenario testing to explore the impacts of varied inputs, model formulations, or solver options. Acceptance tests are useful for comparing model changes or solver performance. There are also online testing tools for more sophisticated challenges.
Nextmv has all the building blocks in place to streamline answering a multitude of questions from challenge work through to judging: Did adding a constraint degrade performance? Do we get better results if we run the model longer? How does the model perform under higher volume inputs? What if we change my objective function? Should we try a different solver? Which team produced the best overall solution? You can also go beyond solution value to understand how the model impacts custom metrics (e.g., number of unscheduled workers). Easily customize your experience by creating your own KPIs to use during the challenge.
Build and practice data-driven presentation skills
Challenge-based events are great for collaborating with new people as well as building communication skills geared toward optimization, engineering, and business-oriented audiences looking to better understand and interact with optimization work. “Why did this model formulation perform better than the previous two?” “Do we anticipate this model change will impact our customer delivery KPIs?” “Why did the model run longer than expected?” These are just some of the questions teams may face.
The Nextmv platform is designed to be collaborative and accessible. It also provides stakeholders with a consistent way to engage with model development, experiment results, model observability, and experiment results.
Where to go from here
You’ve seen some highlights for how Nextmv can be a great springboard for driving innovation in the OR and decision science space through hackathons and challenges. The platform Nextmv removes barriers for collaboration, ideation, and implementation for OR projects — and broadens the opportunities for adoption to adjacent disciplines such as machine learning. Nextmv is a natural solution to try out for your next hackathon, code fest, or innovation challenge. It provides a workflow that fosters teamwork, seamlessly integrates into modern software practices, and is a launchpad for stakeholder buy-in.
If you’re interested in trying out Nextmv for a hackathon, the best way to get started is via a free account. Simply spin one up and start exploring. From there, trial an Innovator plan to get access to adding custom models, inviting collaborators, and executing more runs. This is a great next step for seeing how a larger challenge can come together. You can continue to explore from there or reach out to our team for questions and recommendations.
.png)
