▣ Pushing Capital · Investor Data RoomSession VRN-CFCC40·Viewer investor·2026-05-25 20:23 UTC· NDA locked

Pushing Capital · Investor Brief

Seed Round Data Room

Architecture

How Pushing Capital is built

Production stack on Google Cloud. Same Ingest → Orchestrate → Persist → Serve pattern across every service family. Real today, not aspirational.

The four-verb pattern

1

Ingest

Pull customer data + signals from every source (banks, CRMs, forms, docs, vendor APIs)

2

Orchestrate

Agentic AI workforce + human subcontractors collaborate on the work

3

Persist

Unified golden record in AlloyDB. One source of truth across all service families

4

Serve

Workspace for operators + outcomes for customers. Monetize both sides

Production stack

Compute

  • Cloud Run

    47 services across orchestration, ingestion, APIs, agents

  • Cloud Functions

    Event-driven small tasks (3 functions)

  • Cloud Build + Artifact Registry

    Build/deploy pipeline

Data

  • AlloyDB pc-sovereign-db

    POSTGRES_17 — golden record + RBAC profiles (CEO, Charles, Antigravity, P)

  • BigQuery pushing_capital_warehouse

    Unified financial transaction lake (Mercury · Chase · Stripe · US Bank historical)

  • GCS — 31 buckets

    Bronze/silver/gold medallion data lake pattern

AI

  • Vertex AI Gemini

    On-call CFO assistant + ask-the-data chat (Finance Hub)

  • Named agent identities

    Charles · Antigravity · P + 9+ active A2A agent specialists

  • Subcontractor delivery model

    Licensed accountants, lenders, DMV/IRS specialists consuming golden record

Edge & integrations

  • Cloudflare Workers

    pushingcap-web-v2 (OpenNext) + pc-mcp-platform (webhooks.pushingcap.com)

  • Identity-Aware Proxy

    Google Workspace SSO in front of finance-preview.pushingcap.com

  • Connected systems

    QuickBooks (read-only), Mercury, Chase, Stripe, Gmail signals

Why GCP + AlloyDB (not generic SaaS)

PC's value compounds on the golden record. That requires real database control, real RBAC, real data residency — not the constraints of building on top of someone else's SaaS. AlloyDB gives us Postgres semantics with Google-scale operations. BigQuery handles the analytical workloads (this brief is queried live from there). Vertex AI runs in the same project, no data egress. Consolidating onto one production environment is funded in the seed use-of-funds — moving from today's working setup to a single-tenant, tightly-governed target state.

This investor brief itself is the proof

You are looking at a Next.js app on Cloud Run, querying live BigQuery, served behind IAP, with PC's real bank-statement data flowing through. Built in the same week as the deck. The operator-facing Finance Hub is the same code base. This is the pattern — every PC vertical is an instance.