Laundry

Clean your cross-border payments. Save 70% on FX fees.

Laundry is an AI-powered treasury optimizer that analyzes batches of cross-border payments and routes them through stablecoins to eliminate unnecessary FX conversions. Load your invoices, let the AI find the cheapest path for each one, run a compliance check, and execute — all in one wash cycle. On a $207K batch, Laundry cuts payment costs by 75%.

https://laundry-silk-pi.vercel.app · Getting Started · How It Works · Architecture · Demo

Watch the Demo

---

Table of Contents

• The Problem
• The Solution
• How We're Different
• Getting Started
• Architecture
• How It Works
• Tech Stack
• Configuration
• Demo
• With More Time
• Team
• License

---

The Problem

Cross-border payments are broken. Not because the rails are slow — but because the conversions are wasteful.

When a company pays an international vendor, money typically passes through multiple banks, FX desks, and intermediaries. Each one adds a spread (0.5–5%) and a flat fee ($5–$75). On a batch of 5 invoices totaling $207,000, a company can lose $7,600+ before a single dollar reaches the recipient.

The core issue isn't that transfers are expensive — it's that value is lost repeatedly during conversions and routing decisions that are made without optimization.

The Solution

Laundry is an AI-driven treasury optimizer that minimizes FX losses by reducing or eliminating unnecessary currency conversions.

1. Load — Upload a batch of invoices or payment requests
2. Wash — AI examines the full set of payments, compares traditional bank routing costs against stablecoin routes
3. Spin — Review the optimized routing plan with per-invoice recommendations
4. Rinse — Execute real stablecoin transfers via Crossmint, with full transaction confirmation
5. Brief — AI-generated voice summary of your treasury optimization results

Result: A batch that would lose $7,600 through banks costs $2,070 through Laundry. That's $5,530 saved — a 72% reduction in payment costs.

How We're Different

	Traditional Banks	Fintech (Wise, Crebit)	Laundry
Approach	Fixed FX spreads + fees per transfer	Cheaper FX rates on individual transfers	AI optimizes the entire batch — eliminates conversions, not just cheapens them
Unit of optimization	Single payment	Single payment	All payments together
Intelligence	None	None	AI analyzes flows, recommends routing, executes
Stablecoin usage	None	Stablecoins as cheaper pipe	Stablecoins as conversion elimination — convert once, settle directly
Target user	Everyone	Consumers, students (B2C)	CFOs, treasury teams, companies (B2B)

The key insight: Competitors make the pipe cheaper. We reduce how often you use the pipe.

---

🚀 Getting Started

# Clone the repo
git clone https://github.com/UJameel/Laundry.git
cd Laundry

# Install dependencies
npm install

# Set up environment variables
cp .env.example .env
# Edit .env with your API keys (see Configuration below)

# Start the dev server
npm run dev

Open http://localhost:5173 and you're in.

Other commands

npm run build        # Production build
npm run preview      # Preview production build
npm run test         # Run tests
npm run test:watch   # Watch mode
npm run lint         # Lint with ESLint

---

🏗 Architecture

graph TB
    subgraph Frontend["Frontend (Lovable + React)"]
        LP[Load Page<br/>Invoice batch view]
        WP[Wash Page<br/>AI routing analysis]
        SP[Spin Page<br/>Review optimized plan]
        RP[Rinse Page<br/>Execute transfers]
        BP[Briefing Page<br/>Voice summary]
    end

    subgraph n8n["n8n (Orchestration)"]
        W1[/webhook/laundry-analyze/]
        W2[/webhook/laundry-execute/]
        W3[/webhook/laundry-briefing/]
    end

    subgraph Services["External Services"]
        AI[GPT-4o<br/>Route optimization]
        CM[Crossmint<br/>USDC transfers]
        MM[MiniMax<br/>Voice TTS]
    end

    LP --> WP
    WP -->|Invoice JSON| W1
    W1 --> AI
    AI -->|Route recommendations| WP
    WP --> SP
    SP --> RP
    RP -->|Execute routes| W2
    W2 --> CM
    CM -->|Tx confirmation| RP
    RP --> BP
    BP -->|Summary text| W3
    W3 --> MM
    MM -->|Audio| BP

    style Frontend fill:#1a1a2e,stroke:#6c5ce7,color:#dfe6e9
    style n8n fill:#1a1a2e,stroke:#ea4b71,color:#dfe6e9
    style Services fill:#1a1a2e,stroke:#00cec9,color:#dfe6e9

Loading

---

⚙ How It Works

Data Flow

1. Load — Dashboard loads pre-loaded invoice batch (5 invoices across 5 currencies)
2. Wash — Frontend calls /webhook/laundry-analyze with invoice JSON → GPT-4o analyzes the batch, compares traditional bank routes vs stablecoin routes → returns per-invoice recommendations with reasoning
3. Spin — User reviews the optimized routing plan: cost comparisons, savings per invoice, total batch savings
4. Rinse — User clicks Execute → frontend calls /webhook/laundry-execute → Crossmint transfers USDC on-chain → transaction confirmation returned
5. Brief — AI-generated voice summary via MiniMax TTS plays the treasury optimization report

AI Routing Logic

The GPT-4o agent receives each invoice with:

• Amount in USD
• Target currency and country
• Traditional FX spread (1–6% depending on corridor)
• Traditional flat wire fee ($25–$75)

It compares this against the stablecoin route (1% on-ramp fee, near-zero transfer cost) and recommends the cheapest compliant path for each payment.

---

🛠 Tech Stack

Layer	Technology	Purpose
Frontend	Lovable + React + Vite + Tailwind	Treasury dashboard — invoice view, routing visualization, savings display
Orchestration	n8n	Backend workflow — connects AI analysis, payment execution, and voice generation
AI	GPT-4o (via n8n)	Routing intelligence — analyzes invoices, compares routes, recommends optimal path
Payments	Crossmint	Stablecoin infrastructure — agent wallets, USDC on-ramp, transfers
Voice	MiniMax	Text-to-speech — AI-generated treasury briefings
Deployment	Vercel	Auto-deploys from GitHub on every push

---

🔧 Configuration

Environment variables

Create a .env file in the project root (see .env.example):

# n8n Webhook URLs
VITE_N8N_OPTIMIZE_WEBHOOK=https://your-instance.app.n8n.cloud/webhook/laundry-analyze
VITE_N8N_EXECUTE_WEBHOOK=https://your-instance.app.n8n.cloud/webhook/laundry-execute
VITE_N8N_BRIEFING_WEBHOOK=https://your-instance.app.n8n.cloud/webhook/laundry-briefing

# Crossmint (staging)
CROSSMINT_API_KEY=sk_staging_...

# MiniMax
MINIMAX_API_KEY=...

# OpenAI (used by n8n AI Agent)
OPENAI_API_KEY=sk-proj-...

# Anthropic (optional — billing issues, using OpenAI instead)
ANTHROPIC_API_KEY=sk-ant-...

The VITE_* prefixed variables are used by the frontend. The rest are configured directly in n8n credential nodes.

n8n Workflow setup

The n8n workflow (ID: 3ZQ1rUxmixxFHyRD) contains 3 webhook flows:

1. Analyze (/webhook/laundry-analyze) — Receives invoice batch → GPT-4o AI Agent with Structured Output Parser → Returns routing recommendations
2. Execute (/webhook/laundry-execute) — Receives optimized routes → Crossmint USDC Transfer HTTP Request → Returns transaction confirmation
3. Briefing (/webhook/laundry-briefing) — Receives summary text → MiniMax TTS HTTP Request → Returns audio data

Each flow uses the Webhook → Process → Respond to Webhook pattern.

Required n8n credentials:

• OpenAI API key (for the AI Agent node)
• Crossmint API key (as X-API-KEY header in HTTP Request)
• MiniMax API key (as Authorization: Bearer in HTTP Request)

---

🎯 Demo

Sample scenario: A US-based company has 5 pending international payments:

Vendor	Country	Amount	Traditional Cost	Laundry Cost
Buenos Aires Design Co	Argentina	$12,000	$645 (5%)	$120 (1%)
Berlin Tech GmbH	Germany	$45,000	$1,375 (3%)	$450 (1%)
Lagos Logistics Ltd	Nigeria	$8,000	$555 (6%)	$80 (1%)
London Creative Agency	UK	$92,000	$1,865 (2%)	$920 (1%)
Tokyo Manufacturing	Japan	$50,000	$1,530 (3%)	$500 (1%)
Total		$207,000	$5,970	$2,070

Savings: ~$3,900 (65% reduction) — calculated by AI per batch

The AI analyzes each corridor independently, so actual savings vary by batch composition and FX conditions.

---

With More Time

• Flow netting — Match incoming and outgoing payments in the same currency to eliminate conversions entirely
• Live FX rates — Real-time rate comparison across bank, fintech, and stablecoin routes
• Off-ramp execution — Convert USDC back to local currency as infrastructure expands
• Invoice OCR — Drop a PDF, AI extracts payment details automatically
• Historical analytics — "Last quarter you lost $142,000 to unnecessary FX conversions"
• Multi-company treasury — Optimize flows across subsidiaries of a multinational

---

Team

Name	GitHub	Role
Usman Jameel	@UJameel	Backend + AI orchestration (n8n, MiniMax)
Sritej Bommaraju	@SritejBommaraju	Frontend + Design (Lovable, Figma)
Taka Shirasu	@taka-shirasu	Crossmint integration + Pitch + Animations

---

Built At

Vibe-Coding HACathon — Building AI for Finance Hosted by Hanwha AI Center (HAC) + AI Valley April 3, 2026 | San Francisco, CA

---

License

MIT

---

_{Testnet only — we are not actually laundering anything.}