Aurum

Aurum is an AI-powered family office platform that combines agentic engineering with deep domain expertise to modernize how wealth managers and family offices operate.

The current release delivers the Spend & Oversight Module — transaction flagging, receipt matching, disposition review, and reporting with full audit trails. See the roadmap for what's next.

Current capabilities

• Role-based access (reviewer, manager, admin)
• Dashboard KPIs and action queues
• Flag review queue and disposition workflow with audit trail
• Receipt match queue and confirm/reject workflow with audit trail
• Reports endpoints (JSON + CSV)
• Admin user management

Roadmap

Spend & Oversight Module — live

The current release delivers the core spend analysis and fraud oversight workflow: transaction flagging, receipt matching, disposition review, and reporting with full audit trails.

AI Document Processing — next

Auto-extract and normalize data from bank statements, K-1s, capital call notices, and other financial documents to eliminate manual data entry.

Natural Language Analytics — planned

Query portfolio and transaction data in plain English. "Show me IRR on all deals from 2020-2023" or "Which cardholders exceeded their baseline last quarter?"

Smart Alerts — planned

Proactive notifications for unusual transactions, portfolio rebalancing opportunities, and tax-loss harvesting windows.

Report Generation — planned

On-demand narrative reports from live data. "Create Q4 performance summary for the family" with automatic chart and table generation.

Meeting Prep Assistant — planned

Automatically compile relevant holdings, recent activity, and open items before investment committee or client meetings.

Advanced Capabilities — future

• Predictive cash flow modeling
• Investment opportunity matching
• Risk scenario analysis
• Automated compliance monitoring

Tech Stack

• Next.js 15 (App Router)
• React 19
• TypeScript
• Prisma
• NextAuth (credentials provider)
• SQLite (local default)

Prerequisites

• Node.js 22+
• npm 10+

Getting Started

1. Install dependencies:

npm install

2. Configure environment:

copy .env.local.example .env.local

3. Generate Prisma client and apply migrations:

npm run prisma:generate
npm run prisma:migrate -- --name init

4. Seed local data:

npm run prisma:seed

5. Run the app:

npm run dev

Default URL: http://localhost:3000

Seeded Login Accounts

After running npm run prisma:seed, three dev accounts are created:

• reviewer@example.com
• manager@example.com
• admin@example.com

See prisma/seed.ts for details.

Environment Variables

DATABASE_URL
Local default uses SQLite:

DATABASE_URL="file:./dev.db"

NEXTAUTH_SECRET
Set to a random secret for local/dev/prod.

NEXTAUTH_URL
Example:

NEXTAUTH_URL="http://localhost:3000"

Available Scripts

• npm run dev - start local dev server
• npm run build - production build
• npm run start - run production server
• npm run lint - lint checks
• npm run prisma:generate - generate Prisma client
• npm run prisma:migrate - apply Prisma migrations
• npm run prisma:seed - seed database

Project Structure

app/
  (auth)/login/
  (portal)/
    dashboard/
    flags/
    receipts/
    reports/
    admin/
  api/
components/
lib/
prisma/
types/
docs/

Architecture Diagram

flowchart LR
  U[User Browser] -->|Login + Session Cookies| A[Next.js App Router]
  A --> P[Portal Pages]
  A --> API[API Routes]
  API --> AUTH[NextAuth + AuthZ]
  API --> ORM[Prisma Client]
  ORM --> DB[(SQLite / Postgres)]
  DB --> ORM
  ORM --> API
  API --> P

Loading

Data Model Diagram

erDiagram
  USER ||--o{ FLAG_REVIEW : creates
  USER ||--o{ MATCH_REVIEW : creates
  USER ||--o{ NOTE : creates
  USER ||--o{ MATCH : reviews

  TRANSACTION ||--o{ FLAG : has
  TRANSACTION ||--o{ MATCH : has
  TRANSACTION ||--o{ NOTE : has

  RECEIPT ||--o{ MATCH : links
  FLAG ||--o{ FLAG_REVIEW : audited_by
  MATCH ||--o{ MATCH_REVIEW : audited_by
  NOTE ||--o{ NOTE : replies_to

Loading

Database Objects

• User: authenticated application users with role and status.
• Transaction: normalized card/GL spend transactions and source metadata.
• Flag: suspicious transaction record with risk score/level and lifecycle status.
• Receipt: OCR-parsed receipt metadata and extracted values.
• Match: transaction-to-receipt matching result and review status.
• FlagReview: audit log for reviewer actions on flags.
• MatchReview: audit log for reviewer actions on match decisions.
• Note: threaded investigation comments per transaction.

How Objects Fit Together

• Transaction is the central object.
• Transaction -> Flag supports investigation queueing.
• Transaction -> Match <- Receipt supports receipt reconciliation.
• Flag -> FlagReview and Match -> MatchReview capture immutable decision history.
• User is linked to all authored reviews/notes and optional match review ownership.
• Note supports parent/child replies for simple discussion threads.

Main Routes

Pages:

• /login
• /dashboard
• /flags
• /flags/:id
• /receipts
• /receipts/:id
• /reports
• /admin

APIs:

• /api/auth/[...nextauth]
• /api/dashboard
• /api/flags
• /api/flags/:id
• /api/flags/:id/review
• /api/matches
• /api/matches/:id
• /api/matches/:id/confirm
• /api/notes
• /api/reports/match-summary
• /api/reports/unmatched
• /api/reports/dispositions
• /api/admin/users
• /api/admin/users/:id

Roles and Access

• reviewer: queues, review actions, no reports/admin
• manager: reviewer access + reports
• admin: manager access + user administration

Build and Run (Production)

npm run build
npm run start

Notes

• This project is under active development.
• Local DB defaults to SQLite for ease of setup. To use Postgres, update prisma/schema.prisma datasource provider and DATABASE_URL.
• See docs/TECH_SPEC.md for detailed MVP scope and architecture.

License

MIT