Inferix — India-Aware Semantic Data Type Inferencer

Automatically detect what your CSV columns actually mean — PAN numbers, GST numbers, Aadhaar, IFSC codes, Indian mobile numbers, and 14 more types.

What It Does

Most tools like pandas only detect basic types (int, float, string). Inferix goes deeper and detects the semantic meaning of each column.

Supported Semantic Types (20)

India-Specific (7)

General (13)

Installation

pip install inferix-py

Or install from source:

git clone https://github.com/rakshakr2006-droid/inferix.git
cd inferix
pip install -e .

Quick Start

Train the Model (one-time setup)

python -m inferix.train

Use It

import pandas as pd
from inferix import infer

df = pd.read_csv("your_data.csv")
results = infer(df)
print(results)

Output:

  column_name   semantic_type   confidence  evidence
  cust_pan      pan_number      94%         regex_pan=0.94, name_match=yes
  gst_code      gst_number      97%         regex_gst=0.97, name_match=yes
  join_date     date_disguised  88%         all_int=True, name_match=yes
  monthly_amt   inr_currency    91%         regex_inr=0.85, name_match=yes

Architecture

Inferix uses a 5-layer analysis pipeline combining regex pattern matching, statistical profiling, and XGBoost classification:

Column Data --> [Layer 1: Syntactic] --> null%, unique%, dtype
             --> [Layer 2: Pattern]  --> regex match scores (12 patterns)
             --> [Layer 3: Stats]    --> mean, std, skew, entropy
             --> [Layer 4: Name]     --> column name keyword match
                           |
              [All 50 features combined]
                           |
              [Layer 5: XGBoost Classifier]
                           |
              Semantic Type + Confidence + Evidence

Project Structure

inferix/
├── inferix/
│   ├── __init__.py          # Package init, public API
│   ├── infer.py             # Main infer() function
│   ├── patterns.py          # 12 Indian regex patterns
│   ├── features.py          # 50-feature extraction pipeline
│   ├── data_generator.py    # Synthetic training data
│   ├── train.py             # Model training script
│   └── model/
│       ├── inferix_model.json
│       └── label_encoder.pkl
├── tests/
│   ├── test_patterns.py     # 35 pattern tests
│   ├── test_features.py     # 19 feature tests
│   └── test_infer.py        # 8 inference tests
├── pyproject.toml
├── requirements.txt
└── README.md

Running Tests

# Install dev dependencies
pip install -e ".[dev]"

# Run all tests
python -m pytest tests/ -v

System Requirements

• Python 3.10+
• 8GB RAM (no GPU needed)
• Works completely offline after initial setup

Why Inferix?

License

MIT License

Acknowledgements

Inspired by Sherlock (MIT, 2019) which detects 78 generic semantic types. Inferix fills the gap for India-specific types that Sherlock cannot detect.