AI-Powered Invoice Risk Scoring & Intelligent Payment Reminders
PhantomPay is an AI-driven SaaS prototype designed to help businesses reduce late payments by predicting invoice risk and sending smart, personalized payment reminders. Instead of static follow-ups, PhantomPay adapts its strategy based on customer behavior, invoice history, and reinforcement learning feedback.
This project was built as both a technical exploration of applied AI and a product-focused MVP suitable for early-stage validation, hackathons, and accelerators.
🚀 Key Features
Invoice Risk Scoring
Predicts the likelihood of late or missed payments
Uses historical invoice behavior and metadata
Reinforcement Learning Reminder Strategy
Multi-armed bandit (ε-greedy) approach
Learns optimal reminder timing and tone
LLM-Generated Payment Emails
Personalized, context-aware reminders
Generated using GPT-4o-mini
Explainable Decision Logic
Transparent scoring and reminder selection
Designed for business trust and auditability
Modular SaaS-Ready Architecture
Easily extensible to dashboards, CRMs, and billing systems
🧠 How It Works (High Level)
Invoice Ingestion
Invoice data is collected (amount, due date, customer history, etc.)
Risk Scoring
ML model assigns a risk score (low → high)
Strategy Selection
A reinforcement learning agent selects a reminder strategy
Email Generation
LLM generates a tailored reminder email
Feedback Loop
Payment outcome updates the agent’s learning policy
🏗️ Tech Stack
Backend
Python
FastAPI
SQL (MySQL / SQLite for prototyping)
AI / ML
GPT-4o-mini (email generation)
Reinforcement Learning (ε-greedy multi-armed bandit)
Classical ML models (logistic regression / tree-based experiments)
Data
Structured invoice schemas
Feature-engineered payment history
Dev & Prototyping
VS Code / Replit
GitHub
Hackathon-friendly architecture
📊 Example Use Cases
Small businesses chasing overdue invoices
Finance teams prioritizing follow-ups
SaaS billing platforms adding AI-powered collections
Hackathon demos showcasing applied RL + LLMs
🧪 Current Status
✅ MVP prototype
✅ Risk scoring + RL logic implemented
✅ LLM email generation integrated
🔄 Dashboard & production hardening (in progress)
🔮 Future Improvements
Customer segmentation & clustering
Time-series payment forecasting
CRM integrations (Stripe, QuickBooks, Xero)
Fine-tuned LLMs for tone control
Multi-tenant SaaS deployment
A/B testing framework for reminder strategies
👤 Author
Viplav Dodeja Undergraduate CS Student | AI/ML & Product Engineering 📍 San Francisco Bay Area
📜 License
This project is released under the MIT License. Feel free to use, fork, and build upon it.