Donating is simple. Doing careful nonprofit research is often not.
For donors and advisors, the problem is not access to information. It is that the data is spread across many sources and hard to compare. Ratings can be a starting point, but they rarely explain how an organization operates. And what are they rated against, when both a volleyball team and a 1B hospital have the same rating, what does that mean.
The questions are basic. What does this nonprofit do, who does it serve, and is it making an impact with the dollars it receives?
Many of those answers are in IRS Form 990s. The filings are public, but long and difficult to compare at scale. Financial details are spread across many fields, and program descriptions sit in narrative text that is not easily searchable.
Givespark applies large language models (LLMs) to turn filings and related public information into structured records people can search and compare, with an emphasis on traceability and source verification.
We spoke with Alex, an engineering leader with two decades of experience building large-scale systems, about what makes nonprofit data difficult, how LLMs help, and what matters when AI supports giving decisions.
Q: What is Givespark, and what problem are you solving?
Alex: Givespark helps people discover and evaluate nonprofits.
There are almost 2 million tax-exempt organizations in the U.S., but the data is hard to use. Form 990s are public, yet program descriptions, mission language, and accomplishments are buried in narrative fields that are not searchable at scale.
The financial details are also spread across many sections, which makes comparisons time-consuming.
Most existing tools are built for institutions. We are building something that makes this information easier for individual donors and advisors to use.
Q: What makes nonprofit data so hard to work with?
Alex: The difficulty comes from pulling information from many places and keeping it consistent.
A big part of the problem is that the most important details are often written as free text. Even in Form 990s, the pieces that explain what a nonprofit actually does, its programs, accomplishments, and how it describes its work, sit in narrative fields or just “Additional Notes” pages without any structure at all. That content is valuable, but it is not standardized, which makes comparison across organizations slow and error-prone.
On top of that, IRS filings, state records, watchdog platforms, grant databases, and nonprofit websites often describe the same organizations differently. Doing nonprofit data aggregation well means reconciling those differences without losing context.
Q: Why focus on form 990s?
Alex: We needed a good point to start from. 990 forms are good as they are structured and contain a lot of information previously hard to reach at scale.
They include financial reporting, but also narrative sections where nonprofits describe their mission, programs, and accomplishments. That content matters because it describes the organization’s work in its own words.
The challenge is scale. It is hard to compare hundreds of filings when key information is embedded in unstructured text. LLMs make it possible to label and structure that narrative so it can be searched and compared.
Q: How does the AI work in practice?
Alex: We start with IRS Form 990 filings and add other public information when it provides context. The goal is to turn hard-to-compare text into structured fields so people can search and compare organizations more easily.
Much of what donors care about is written as free text. In Form 990s, that includes mission statements, program descriptions, and accomplishments. Similar details can appear on nonprofit websites. This information is useful, but it is not standardized, so comparing organizations takes time.
We use AI to extract consistent details and organize them into categories like cause area, who is served, geography, and program approach. We also keep the source visible. When a summary or label is shown, users can view the original wording it was based on.
One limitation is timing. Form 990s are typically annual, so they can be accurate but not current.
To add more recent context, we also analyze nonprofit website updates and are exploring AI-supported tracking of credible news coverage. Any newer information is treated as context and kept separate from IRS filings, so users can clearly see what is officially reported versus what is more recent public information.
Q: What is the hardest technical challenge?
Alex: Connecting organizations and their relationships.
Nonprofits receive grants, work with fiscal sponsors, and operate through chapters or parent entities. The same relationship can appear in multiple places and be reported differently.
That is where entity resolution AI becomes important. You need to determine whether two records refer to the same organization and be able to explain why they were linked.
This also requires multi-source data analysis. No single dataset captures nonprofit relationships accurately, especially when organizations report information differently across filings and public sources.
Q: You built multiple great projects in the fintech space and are now building a product in the non-profit sector. Why move?
Alex: I spent years building systems in regulated environments, including platforms that processed billions in federal loans at high volume. In that context, compliance and explainability shape architecture from the start.
That mindset carries into nonprofit transparency. If an output influences giving decisions, it needs clear audit trails and the ability to explain how results were produced. I believe that my experience from highly lucrative industries like fintech can bring a lot of healthy and long overdue innovations to one of the most important and overlooked sector.
Q: How do you think about trust with ai in the loop?
Alex: Trust comes from provenance and clarity.
People should be able to trace a label or summary back to source material and make their own judgment. They also need to see where the data is coming from and whether it is based on official filings or other context.
That is part of what makes Givespark a philanthropic data platform, not just a search tool. The goal is not to replace due diligence. It is to make it easier to start with better information.
Q: Where is AI heading in philanthropy?
Alex: LLMs make it possible to organize nonprofit information at a scale that was not practical before.
For Givespark, the next step is natural language search. Instead of learning filters, people can describe what they are trying to support and get results backed by source material. This reflects a broader shift in Charitable Giving Technology, where discovery becomes more conversational but still grounded in verifiable data.
What most people actually want is cause-based discovery. They are not searching for a nonprofit by name. They are searching for a specific need, like “after-school tutoring in Chicago” or “wildlife rescue in coastal areas.” Good tools should take that intent, find relevant organizations, and show enough detail to compare them without forcing users to read dozens of filings.
I also expect to see more AI for social impact in grant matching and research workflows. The tools that hold up over time will make it easy to verify where information comes from.
Q: What advice would you give founders building in traditional industries?
Alex: Start with a real problem.
Many industries have information gaps and outdated tooling. Straightforward AI can help if you focus on data quality and user trust.
Be clear about limitations. AI makes mistakes. Systems should show sources, expose uncertainty, and let users verify outputs.
What’s Next For Givespark
Givespark is focused on making nonprofit information easier to access and compare, without assuming users have institutional budgets.
Over time, that includes tools that support nonprofit financial analysis and improve how donors and advisors search by mission and approach. The broader goal is to make nonprofit research easier for people who want to give carefully, whether they are donating personally or advising clients.
About the Interviewee
Alex Bondarevskyi is a software engineer and engineering leader with 20 years of experience building systems that operate at a national scale. During the pandemic, he led engineering for one of the largest non-bank PPP lenders in the country, joining as the first engineering hire and building the team and infrastructure that processed nearly 2 million loan applications worth $12.5 billion, serving 600,000+ independent contractors at peaks of 50,000 applications daily. He previously co-founded ComCard, a fintech startup in corporate payments, and ran a development agency serving major film studios and Fortune 500 companies. Currently, he’s building GiveSpark, a charity discovery platform helping donors and advisors navigate 1.8 million US nonprofits.
