Finance Credit Follow-Up Email Agent

Task 2 · AI Enablement Internship Project

An autonomous AI agent that monitors overdue invoices and sends progressively escalating payment reminder emails — keeping client relationships intact while reducing Days Sales Outstanding (DSO).

---

Features

• Automated Invoice Processing — Reads pending invoices from CSV/Excel
• 4-Stage Tone Escalation — From warm & friendly to stern & urgent
• LLM-Powered Emails — Google Gemini generates personalised, professional emails
• Escalation Flagging — 30+ day records flagged for manual finance/legal review
• Immutable Audit Trail — Every action logged to SQLite with SHA-256 body hashes
• Dry-Run Mode — Safe testing without sending real emails
• Streamlit Dashboard — Visual interface with real-time metrics

---

Quick Start

Prerequisites

• Python 3.11+
• A free Google Gemini API key (Get one here)

Installation

git clone https://github.com/Advitiyyaaa/Finance-Credit-Follow-Up-Email-Agent
pip install -r requirements.txt
cp .env.example .env
# Edit .env with your Google API key

Run the Agent (CLI)

# Dry-run mode (default — no real emails)
python run_agent.py --dry-run

# With custom input file
python run_agent.py --input data/sample_invoices.csv --dry-run

# View audit log
python view_audit.py --last 50

Run the Dashboard (Streamlit)

streamlit run app.py

---

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                        ORCHESTRATOR                             │
│               (LangGraph StateGraph)                            │
└────────────────────────┬────────────────────────────────────────┘
                         │
         ┌───────────────▼───────────────┐
         │       DATA INGESTION          │   CSV / Excel → pandas
         └───────────────┬───────────────┘
                         │
         ┌───────────────▼───────────────┐
         │     OVERDUE CLASSIFIER        │   days_overdue → stage
         └───────────────┬───────────────┘
                         │
         ┌───────────────▼───────────────┐
         │    EMAIL GENERATION           │   Gemini 2.0 Flash (LLM)
         └───────────────┬───────────────┘
                         │
         ┌───────────────▼───────────────┐
         │      SEND / DRY-RUN           │   Console log + JSON
         └───────────────┬───────────────┘
                         │
         ┌───────────────▼───────────────┐
         │        AUDIT LOGGER           │   SQLite audit trail
         └───────────────┬───────────────┘
                         │
         ┌───────────────▼───────────────┐
         │    ESCALATION FLAGGING        │   30+ days → manual review
         └───────────────────────────────┘

---

Tech Stack

Layer	Technology
LLM	Google Gemini 2.0 Flash (free tier)
Agent Framework	LangGraph 0.2.x
Data Ingestion	pandas, openpyxl
Output Validation	Pydantic v2
Audit Log	SQLite
UI	Streamlit
Secrets	python-dotenv + .env

---

Tone Escalation Matrix

Stage	Days Overdue	Tone	CTA
1	1–7	Warm & Friendly	Pay now link
2	8–14	Polite but Firm	Confirm payment date
3	15–21	Formal & Serious	Respond within 48 hrs
4	22–30	Stern & Urgent	Pay immediately
ESC	30+	🚫 No auto-email	Assign to finance manager

---

Security Mitigations

Risk	Mitigation
Prompt Injection	Input sanitisation (HTML stripping, length truncation), structured JSON output, Pydantic validation
Data Privacy / PII	Local processing, email masking in logs, SHA-256 body hashing in audit
API Key Exposure	.env + python-dotenv, .gitignore protection
Hallucination	Post-generation validation comparing LLM output against source data
Duplicate Sends	Audit log check before each send (same invoice_no + stage)
Rate Limiting	Max 100 invoices per run

---

Project Structure

├── agent.md                 # Architecture document
├── run_agent.py             # CLI entry point
├── view_audit.py            # Audit log viewer
├── app.py                   # Streamlit dashboard
├── requirements.txt
├── .env.example
├── src/
│   ├── config.py            # Configuration & constants
│   ├── models.py            # Pydantic data models
│   ├── ingest.py            # CSV/Excel ingestion
│   ├── classifier.py        # Overdue stage classifier
│   ├── email_generator.py   # LLM email generation
│   ├── sender.py            # Send / dry-run
│   ├── audit.py             # SQLite audit trail
│   ├── escalation.py        # Escalation flagging
│   └── graph.py             # LangGraph StateGraph
└── data/
    ├── sample_invoices.csv  # Sample data
    └── (audit_log.db)       # Created at runtime

---