The Opportunity Project: A new model for civic tech
By Nathan Zencey
As part of our public sector work, Civis is always on the lookout for pro bono projects. Not only does this offer us a way to contribute to civic good, but we also get better at solving problems we don’t normally work on.
A new model we really like for this kind of civic tech is the Census Bureau’s “Opportunity Project,” a community-like garden for federal government technology projects. Any federal agency can put out a problem statement, and they in turn receive seed products from various companies who are interested in helping out. The Census Bureau hosts a “Demo Day” in which companies show off their seed products to agencies and other potential partners.
Companies develop their products separately but in parallel, maintaining the rights to their intellectual property. If there’s mutual interest in cultivating a product after the demos, companies and the government agencies can enter into a more formal relationship through procurement. At the very least, the government agencies get to see new ways of approaching difficult problems with civic tech; companies in turn get insight into what problems agencies are trying to solve with data science — not to mention a workable prototype or demo.
We chose to work on a project through the Department of Labor (DOL) to help returning veterans find registered apprenticeships. While there is no single experience for a returning veteran, many veterans, particularly those who have been serving their entire adult lives, don’t know how their skills translate to the civilian world or how to put their experience and accomplishments in a relevant civilian context. DOL wanted a product that could take language input from a veteran — their resume, interests, and Military Occupational Classification (MOC) — and find them the most relevant registered apprenticeships.
When it came to building a product, we found the real power of the Opportunity Project was the quick ramp-up process. Rather than long and protracted negotiations over contract language, DOL introduced us right away to veterans and advocates who gave us valuable advice about what features a potential product should have. And DOL organized the development as a “sprint” — a timeline of major milestones met quickly over the course of several weeks.
The Opportunity Project also helps companies find and build new partnerships that can last well beyond the original product development. We collaborated with an impressive technical partner on this project: SkillsEngine, a non-profit based in Austin, Texas that offers extremely efficient and accurate parsing of resumes and job descriptions. Through their API, SkillsEngine helped us take a massive leap in this project; essentially, they let us start turning letters into numbers right away.
Here’s a look at our product in action:
In this example, a user enters an MOC to help refine results; MOC 91E corresponds to an Allied Trades Specialist (i.e. Machinist) in the Army. As you can see, the algorithm incorporates this additional information to change the results from a mix of machinist and mechanic roles to just machinist apprenticeships.
Our task was to match veterans with apprenticeships, which are fundamentally about learning new skills. In our demo, if a user wants to learn unique skills of a specific trade, they can simply list those skills and receive highly relevant results (given the constraints of the limited database we used for the demo).
However, a much harder use case is helping a veteran find an opportunity based on some but not all skills from a previous profession, especially “soft skills” — i.e., “I never want to do X again — but I do miss leading a small team. I want to see some different types of positions where I can do that.” For this use case, our product’s matching is too “eager” — it’s always providing the most literal matches based on user input, rather than providing “wild card” options that are more loosely connected to past experience. In the future, the simplest way to improve matching outcomes might be through more explicit user input — “I want to learn something new” vs. “I want to apply something I’m already good at,” and then asking for specifics and developing separate matching pipelines and databases for each.
We look forward to participating in Demo Day on March 1, where we’ll share our work and hopefully provide some inspiration for government entities or non-profits looking to do something similar.