MT

Architecture

How the platform works end to end — and exactly where the data is real. Nothing in the UI is hard-coded; every figure is queried live from Postgres.

Database tables

9

Supabase Postgres

Live records

133

queried this request

Sourced signals

31

each with a source URL

Real nonprofits

13

real IRS EINs + revenue

Hosting

Vercel

edge, per-request SSR

Data pipeline

1 · Real sources

Data origin

  • SEC EDGAR — 10-K / 10-Q filings
  • IRS / ProPublica Nonprofit Explorer API
  • Company sustainability reports & press
  • B Lab (B Corp) directory

Every signal keeps its source URL

2 · Ingestion

Scripts

  • fetch_nonprofits.py → live IRS pull
  • Deep research → sourced signals
  • gen_seed.py → deterministic SQL
  • Supabase migrations (schema + seed)

Reproducible, versioned in repo

3 · Supabase Postgres

System of record

  • 9 tables · 133 rows
  • Row-Level Security enabled
  • companies · signals · contacts
  • lots · matches · shipments · tax_records

Live database — counts below are queried now

4 · Next.js on Vercel

Application

  • React Server Components (queries.ts)
  • Reads via anon key + RLS policy
  • Server Actions write via service role
  • Edge-deployed, per-request rendering

No data is hard-coded in the UI

5 · Surfaces

What you see

  • Grow — prospecting pipeline
  • Operate — donation fulfillment
  • Network · Compliance · Impact
  • ⌘K command palette

Every number traces back to the DB

Data provenance — real vs. modeled

DataTypeSource
Company facts (revenue, employees, HQ, ticker) RealSEC filings, annual reports, company sites
Buying signals (write-downs, ESG, B-Corp) RealSEC 10-K/10-Q, press — source URL on each signal
Executive contacts (name, title, role) RealCompany leadership pages / press — no emails fabricated
§170(e)(3) tax eligibility & parent routing RealEntity analysis (US C-corp vs foreign parent)
Nonprofits (name, EIN, location, revenue) RealIRS Form 990 via ProPublica Nonprofit Explorer
Pipeline stage, fit score, excess estimate ModeledModeled — grounded in the real write-down figures
Donation lots, matches, shipments, tax records ModeledModeled demo pipeline (Simpli hasn't run these yet)
Nonprofit capacity & rating ModeledIllustrative

Calculation lineage

Every deduction shown is computed from stored basis & FMV — not entered by hand.

IRC §170(e)(3)

deduction =
min( basis + ½·(FMV − basis),
     2 × basis )

Informational, not tax advice.

Stack

Next.js 16React 19Server ActionsTailwind v4TanStack TableSupabase PostgresRow-Level SecurityVercelcmdk

22 real companies tracked

13 real 501(c)(3) partners