AName
Phi Pham
BWhat I do
I build the back office small nonprofits can't afford to hire: a company brain, finance that closes on time, and data rules staff actually follow.
Tell me what's stuck. I'll read your filings before we talk.
AName
BWhat I do
I build the back office small nonprofits can't afford to hire: a company brain, finance that closes on time, and data rules staff actually follow.
Tell me what's stuck. I'll read your filings before we talk.
For small teams, usually a five-to-twenty-person nonprofit where the executive director is also the IT department, the bookkeeper and the grant writer.
Everything your org knows, from Drive, email and meeting notes, turned into a knowledge base your team can ask questions of inside the AI tools they already use. You own the files. You can walk away with them.
How CazMind works ↗BuiltA 1,583-page, permission-aware brain built from a decade of one nonprofit's records. 98% retrieval accuracy in testing. The engine is open source.
Month-end close, budget against actuals, cash-flow forecasts, payroll and the written procedures that let it survive a staff change.
Running nowFractional finance lead for an arts nonprofit: monthly close, payroll and a cash-flow model rebuild. At Building Beats, a bookkeeper-handoff SOP that went live in September 2026.
What staff can paste into ChatGPT or Claude, what they can't, and why. Student and donor data handled under the laws that actually apply to you.
The notes I work from ↗DeliveredA data guide for a New York youth tech nonprofit: account risks, a de-identification workflow for student records and a one-page AI policy.




Six reads of public 990s and audits, written while pitching. Names removed. These are diagnostics, not client results.
| Line | The organization | What the filings showed | What I proposed |
|---|---|---|---|
| 01 | A ~$10M national fiscal sponsor for open-science projects | $18.9M to $8.7MRestricted funds fell as three projects spun out, leaving 1.2 to 1.8 months of unrestricted reserves. | A 100-day plan. Its three first-year audit findings were all fixable controls. |
| 02 | A ~$540K summer camp in upstate New York | −$190,157Cumulative net result, FY2019 to FY2024, read across fourteen years of returns. Expenses rose 25.1% in one year. | A one-page board dashboard and a reporting calendar. |
| 03 | A ~$240K all-volunteer nonprofit regranting in the DRC | Under one monthOf reserves on hand, while 92% of spending went straight to partners. | Five funders, eligibility checked, sequenced through spring 2027, with the first request drafted. |
| 04 | A ~$270K foundation regranting in South Africa | 3 deficits in 5 yearsAll revenue from contributions, reserves sliding from $605K to $563K. | A 90-day plan to build a monthly-giving base. |
| 05 | A community cinema in New York, mid capital campaign | 40%, at bestThe closest comparable's earned revenue never covered half of expenses in ten years of filings. | A driver-based operating model that separates sourced numbers from assumptions. |
| 06 | A ~$176K refugee aid nonprofit in Middle Tennessee | About three monthsOf operating cash, with a PayPal-only giving page as the bottleneck. | Ten funders, each tagged verified, likely fit or worth checking. |
Every figure comes from public IRS filings or published audits. The full write-ups go to the organization, not this page.
My parents came over from Vietnam with almost nothing and found each other in Boulder. I started DJing in high school, booking my own gigs. That was the first business I ran, and the first time I understood that culture organizes people.
On air with Ernie Anastos.New York local news, 2017.
Small shipped apps. They're how I keep my hands on the tools I recommend.