finee 1.0.0

Extract structured financial entities from Indian banking messages

---

language:

• en license: mit library_name: transformers tags:
• finance
• entity-extraction
• ner
• phi-3
• production
• gguf
• indian-banking
• structured-output base_model: microsoft/Phi-3-mini-4k-instruct pipeline_tag: text-generation

---

Finance Entity Extractor (FinEE) v1.0

A production-ready 3.8B parameter language model optimized for zero-shot financial entity extraction.
Validated on Indian banking syntax (HDFC, ICICI, SBI, Axis, Kotak) with 94.5% field accuracy.

[ Model Card ] · [ GitHub ] · [ Quick Start ]

---

Performance Benchmarks

Comparison with Foundation Models

Model	Parameters	Entity Precision (India)	Latency (CPU)	Cost
FinEE-3.8B (Ours)	3.8B	94.5%	45ms	Free
Llama-3-8B-Instruct	8B	89.4%	120ms	Free
GPT-3.5-Turbo	~175B	94.1%	~500ms	$0.002/1K
GPT-4	~1.7T	96.8%	~800ms	$0.03/1K

Platform Support

Platform	Framework	Status
macOS Apple Silicon	MLX	✅ Full Support
Linux + NVIDIA GPU	PyTorch/Transformers	✅ Full Support
Linux + CPU	PyTorch/GGUF	✅ Full Support
Windows	GGUF/llama.cpp	✅ Full Support

🐍 Quick Start with FinEE Library

The easiest way to use the model is through the finee Python library, which handles backend selection, caching, and validation automatically.

Installation

# Install from GitHub
pip install git+https://github.com/Ranjitbehera0034/Finance-Entity-Extractor.git

# Or clone and install locally
git clone https://github.com/Ranjitbehera0034/Finance-Entity-Extractor.git
cd Finance-Entity-Extractor
pip install -e ".[metal]"   # Apple Silicon
pip install -e ".[cuda]"    # NVIDIA GPU
pip install -e ".[cpu]"     # CPU only

Usage

from finee import extract

# Automatic backend detection (MLX, CUDA, or CPU)
text = "Rs.500 paid to swiggy@ybl on 01-01-2025"
result = extract(text)

print(f"Amount: {result.amount}")
print(f"Merchant: {result.merchant} ({result.category})")
print(f"Confidence: {result.confidence.value}")

# Output JSON
print(result.to_json())
# {
#   "amount": 500.0,
#   "type": "debit",
#   "merchant": "Swiggy",
#   "category": "food",
#   "date": "01-01-2025",
#   ...
# }

Command Line Interface

# Direct extraction
finee extract "Rs.500 debited from A/c 1234"

# Check available backends
finee backends

---

📋 Overview

This project demonstrates how to:

1. Parse 40K+ emails from a Gmail MBOX export
2. Classify emails into categories using Phi-3 Mini
3. Discover patterns in financial emails (transactions, amounts, dates)
4. Fine-tune a local LLM using LoRA for entity extraction
5. Extract structured data: amount, transaction type, account, date, reference

🏗️ Project Structure

Finance-Entity-Extractor/
├── src/
│   └── finee/                 # FinEE Package
│       ├── __init__.py
│       ├── extractor.py       # Main pipeline orchestrator
│       ├── cache.py           # Tier 0 LRU Cache
│       ├── regex_engine.py    # Tier 1 Regex Engine
│       ├── merchants.py       # Tier 2 Rule Mapping
│       ├── prompt.py          # Tier 3 Targeted Prompts
│       ├── validator.py       # Tier 4 Validation & Repair
│       ├── backends/          # Auto-detecting Backends (MLX, PT, GGUF)
│       └── cli.py             # Command Line Interface
├── tests/                     # 88 Unit Tests
├── .github/workflows/         # CI/CD
├── pyproject.toml
├── train.py                   # Training pipeline
└── README.md

🎯 Extracted Entities

Entity	Description	Example
amount	Transaction amount	"2500.00"
type	Debit or Credit	"debit"
account	Account identifier	"3545"
date	Transaction date	"28-12-25"
reference	UPI/NEFT reference	"534567891234"
merchant	Merchant name	"swiggy"
category	Transaction category	"food"
confidence	Extraction confidence	"HIGH"

📈 Benchmark Results

Multi-Bank Validation (v8)

Bank	Field Accuracy	Status
ICICI	96.2%	✅
HDFC	95.0%	✅
SBI	93.3%	✅
Axis	93.3%	✅
Kotak	92.0%	✅
Overall	94.5%	✅

Field-Level Accuracy

Field	Accuracy
Amount	98.5%
Type	99.2%
Date	97.8%
Account	96.1%
Reference	72.7%

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

• Microsoft for Phi-3 model
• MLX team for the amazing framework
• Hugging Face for model hosting

---

Made with ❤️ by Ranjit Behera