You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
A free, searchable tool that makes the government's own reported improper payments legible to any journalist, watchdog, or citizen. The government reports roughly $186 billion a year in improper federal payments (GAO, FY2025), over $3 trillion cumulatively since FY2003, yet the public can do nothing with that number. The raw data that would expose the coordinated patterns (straw owners, excluded providers still billing, duplicate draws) is public but scattered across a dozen incompatible portals no citizen can join. I already run this forensic analysis privately across Medicare claims, PPP loan records, and federal exclusion lists. The problem is not missing data, it is missing legibility. This grant turns a private capability into public infrastructure.
Goal: make federal payment integrity data legible to the public, starting with one program and extending outward. The v1 deliverable, within 90 days of funding, is a single live, public, searchable view of one federal program's improper-payment and excluded-provider exposure. Real data, real cross-dataset joins, a working URL anyone can use, architected to extend to the next program, then state Medicaid and federal contracts. The pipelines already run. LLMs now make ten-million-row joins and messy-schema reconciliation tractable for one person, which is what makes a solo, AI-native build possible. I am not proposing to catch fraud for the government. I am making the government's own leakage legible to everyone else and letting daylight do the rest.
$25,000 goal, with an $8,000 minimum viable. This is not salary. It buys public delivery of an existing private capability: compute and LLM inference for the large cross-dataset joins, data acquisition and a regular refresh cadence, hosting and a public front end, and part-time runway to ship v1 within 90 days. At the $8,000 minimum I ship a single-program public view. The full $25,000 funds a faster build, a second program, and the refresh cadence that keeps it current.
Solo: Joshua Elberg, self-taught forensic data analyst and founder of Palavir LLC (Detroit). I run fraud and payment-integrity analysis across federal datasets at a scale most people assume only a government inspector general can touch: Medicare Part B and Part D claims, PPP loan records, and federal exclusion lists, joined across tens of millions of rows to find the patterns that sampling-based audits miss. No citizen-facing, cross-program version of this exists today. The tools live inside IG offices and a few contractors. The pipelines already run. I use them in my own analysis today. (Figures FY2025, primary-source verified: GAO GAO-26-108694, DOJ OPA, CMS FY2025 improper-payments fact sheet.)
The most likely failure mode is a data-access or refresh obstacle on the first program: a portal changes format, a bulk file gets rate-limited, or a source moves behind a new login. Because the pipelines already run privately, the risk is not whether the analysis works, it is whether the public delivery layer stays current. Mitigation: I start with the program whose source data is most stable and openly downloadable, and I scope v1 to a single program so the surface area stays small. The worst realistic outcome is that v1 covers one program instead of two, which is still a working public tool that did not exist before. There is no outcome where the money is spent and nothing ships, because the underlying analysis is already built.
$0 in outside funding for this specific project. Palavir LLC is self-funded. I have a separate application in with Emergent Ventures (Mercatus Center) for the same public tool, submitted in July 2026, with no decision yet. This is my first outside funding request dedicated to the legibility engine.