Supreetha15/pii-safe-classifier

🧠 PII-Safe Email Classifier API

A FastAPI backend that classifies support emails using SBERT + SVM, while safely masking and restoring personally identifiable information (PII) using regex and SpaCy NER.

🔧 Tech Stack

• —FastAPI – Lightweight REST API backend
• —SBERT (MiniLM-L6-v2) – Sentence embeddings
• —Linear SVM – Email classification
• —Regex + SpaCy – PII detection & masking
• —Joblib – Model persistence
• —Uvicorn – ASGI server

🔌 API Overview

• —Framework: FastAPI
• —Main file: api.py
• —Model: SBERT encoder + Linear SVM loaded via models.py
• —PII masking: Done in utils.py using regex & SpaCy
• —Entry point: app.py (used for Hugging Face Spaces or Docker)

☁️ Hugging Face Spaces Deployment

• —Create a Space and choose Docker as the SDK.
• —Upload all project files: app.py, api.py, models.py, utils.py requirements.txt Dockerfile README.md sbert_linear_model.joblib
• —Space will build automatically and give the following link:
• —https://<space-name>.hf.space/docs
• —Use Swagger UI or curl to test live requests.

📚 Module Documentation

1.app.py - Hugging Face Spaces entry point; exposes FastAPI app instance from api.py.

2.api.py - FastAPI route /classify for processing incoming emails: masks PII, classifies, and demasks.

3.models.py - Contains SBERTSVMClassifier class: training, saving, loading, and predicting using SBERT + SVM.

4.utils.py - Implements mask_pii() and demask() using advanced regex and SpaCy for secure entity handling.

5.requirements.txt - Python dependencies required to run the application (locally or in Hugging Face Space).

6.Dockerfile - Containerization file to deploy the API using Hugging Face's Docker SDK or locally with Docker.

7.README.md - Complete setup guide, documentation, deployment steps, and test instructions.

8.sbert_linear_model.joblib - Serialized classifier model file (must be pre-trained and included for prediction).

API End Point Details

POST /classify Content-Type: application/json

Body Schema: { "email_body": "string" }

⚙️ How It Works

1. 1.Accepts raw email text (email_body)
2. 2.Detects and masks PII using regex and NER
3. 3.Classifies the masked email into 1 of 4 categories:
4. 4.Incident, Request, Problem, Change
5. 5.Gives the position of the entity like email position,full_name position etc.
6. 6.Demasks PII for final response
7. 7.Returns output in strict JSON schema

⚙️ How to work(Using Swagger UI)

1. 1.Use this url "https://supreetha15-pii-safe-classifier.hf.space/docs"
2. 2.This will redirect to Swagger UI.
3. 3.Use POST and the route by default will be /classify.
4. 4.Check on Try Out to to evaluate the API bu giving a raw email text in email_body
5. 5.Click execute to run the API and fetch the result.

Other Evaluation methods

1. 1.Use this url "https://supreetha15-pii-safe-classifier.hf.space/classify"
2. 2.Go to POSTMAN and set request to POST
3. 3.In body select raw JSON and provide a raw email text as {"email_body":"string"}
4. 4.Send the request in order to get the API response
5. 5.Other Equivalent method can be using curl.

Acknowledgements

Built with using open-source tools like FastAPI, SBERT, and SpaCy. Deployed on Hugging Face Spaces using Docker SDK.

📜 License

This project is licensed under the MIT License – you're free to use, modify, and distribute.

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference