How college students, volunteers, and advanced AI joined forces to create the Internet’s only search engine for service.

by Oshmi Ghosh, marketing intern, novelist, NYU student

The technological world is changing faster than ever before. Often it seems we, as regular people, cannot catch up.

Artificial intelligence especially appears wherever you look, and it’s a natural instinct for many to view its arrival as encroachment rather than enhancement. Samaritan Scout, an AI-powered search engine for volunteer opportunities, understands your concerns and is here to be completely upfront about the role of AI in our software. Hopefully, we can assuage your fears and take small but meaningful steps toward a more informed understanding of our changing planet.

When Samaritan Scout began, there were a handful of talented software developers involved in setting the foundation for what the platform is today. Dvora Inwood, the co-founder and head of product, approached Will Rosenthal, then a high-school student in Cranford, NJ, who ultimately settled the proof of concept. Without any assistance from AI, Rosenthal worked to program efficient ways to extract relevant data from websites to compile on one platform. At the time, chatbots and all forms of artificial intelligence were unreliable and often sloppy, which left most of the busywork for human hands. 

Soon accompanying Rosenthal were Wyatt Sell and Pradhi Pakkerakari, as well as the first batch of software development interns. Samaritan Scout’s concept was singular; no other search engine promised such quick, streamlined, and accurate location-based list generation. In order to keep this promise, the team had to look for ways to supplement their own skillsets.

From Web Crawlers to ETL Pipelines: How Data Extraction Powers an AI-Powered Search Engine for Volunteering

Then came a considerably more evolved version of ChatGPT. The inclusion of AI streamlined and organized the development process, or ETL. The first step, extraction, runs through pages upon pages to pull data, which started off as slow manual labor but now uses AI to produce larger bodies of information. The second, transformation, sorts through this data based on specifications communicated through prompt engineering. 

The third, loading, brings in the AI’s input which assigns each ‘Volunteer Opportunity’ (VO) accurate classifications and converts parcels of text (such as paragraphs describing the VO) into vectors that we can then compare to other texts. Each time a user searches for a VO on our website, vector embedding scan these classifications and ensure that the search results suit the user based on submitted information and previous searches.

Of course, AI makes mistakes, and any business looking to leverage AI must be aware of these concerns beforehand. The early AI models would hallucinate, or generate entirely made-up information that only matched the prompt in wording or theme. Since our AI models occasionally fabricated locations of VOs, the software team had to then manually aggregate every county and zip code in the US into an index that the model would have to choose from, facilitated by an API (Application Programming Interface) that forced the AI to consult real data from real maps. The model also started including donation pages as potential VOs, meaning that even stronger specifications needed to be made. 

Human Oversight in Artificial Intelligence: Why Volunteers, Interns, and Developers Still Matter More Than the Model

Ultimately, human interference governs Scout's every interaction with AI and will continue to do so even as more sophisticated forms of AI emerge. The model may have been improved by artificial intelligence, but it was not created by artificial intelligence and could never have been. Our incredibly skilled volunteers, interns, and supervisors have collaborated to push Samaritan Scout ahead of the tech race using their own ingenuity.

“The whole point of Scout is to help and to connect,” says Rosenthal, “and we need to make sure that people can do that.”

Our organization has always been premised on human connection and charitable values, which means that real people are on the frontlines building these connections and espousing these values at every step. Our computers engender efficiency; they do not take control of our unique skillsets.

Right now, Akash Dubey and Jack Greenwald work to leverage the more sophisticated versions of AI available to the team. Every time a new issue arises, such as ineligible VOs or barriers preventing us from welcoming a new U.S. state, they spend hours refining the prompt to make sure AI remains a reliable resource. They have to adjust the parameters as they switch between language models to limit the possibility of hallucination and deliver relevant results on our website.

“[This process] makes sure that what the user asks is what they get,” assures Dubey.

Additionally, nothing on the platform itself, besides the list of VOs, is AI-generated. All of our writing, outreach, social media, and design-work are conceived, done, and reviewed by actual people. If you use our platform, not only will you be engaging with hard work and true dedication, you will also gain valuable human experience in whatever volunteer work you find through our website. Scout covets real presence and teamwork, as should all companies investing in a strong, authentic world. 

The Future of AI-Powered Data, Search, and Volunteering: Building Innovation Without Losing Human Values

To those of you still nervous about AI’s growth over the years, we truly understand. We are, therefore, fully dedicated to identifying the appropriate roles that AI can play. AI is only one small piece in a large puzzle. What we really want to do is innovate, fulfill people’s dreams, and make lives easier. In hands like ours, AI is not something to be feared. Together, we can figure out how to navigate these complicated times and build a brighter, more collaborative future.