SecureSum

Privacy-first AI meeting and transcript summarizer that redacts sensitive data before AI processing.

SecureSum is a full-stack application designed to reduce context leakage risks when using AI for meeting notes, transcripts, research logs, or internal documents. Instead of sending raw text directly to an LLM, SecureSum first detects and masks sensitive information, then generates summaries and stores safe history for future reference.

---

Why This Project Matters

Most AI summarization tools process raw content directly, which can expose:

• emails
• phone numbers
• API keys
• card-like numbers
• internal identifiers
• confidential notes

SecureSum introduces a privacy layer before summarization.

---

Core Features

Privacy Layer

• Detects sensitive patterns
• Masks confidential data before processing
• Returns sanitized output

AI Summaries

• Generates concise summaries
• Extracts structured insights
• Supports future upgrade to Gemini / Groq / OpenAI

Analysis History

• Stores previous analyses in MongoDB
• Enables future search and audit workflows

Full-Stack Architecture

• Modern frontend
• REST API backend
• Python ML microservice
• NoSQL persistence

---

Tech Stack

Frontend

• Next.js
• React
• Tailwind CSS
• TypeScript
• Axios

Backend

• Node.js
• Express.js
• Helmet
• Morgan
• CORS

ML Service

• FastAPI
• Python
• Regex-based PII detection

Database

• MongoDB

---

System Architecture

User Input
   ↓
Next.js Frontend
   ↓
Express Backend API
   ↓
FastAPI Sanitization Service
   ↓
Summary Generation
   ↓
MongoDB Storage
   ↓
Dashboard History

---

Current Workflow

1. User pastes transcript or notes
2. Frontend sends text to backend
3. Backend calls sanitization service
4. Sensitive data is masked
5. Summary is generated
6. Result is saved to MongoDB
7. User sees history dashboard

---

API Endpoints

POST /api/analyze

Analyze text input.

Request

{
  "text": "Meeting transcript here"
}

Response

{
  "id": "...",
  "sanitized_text": "...",
  "summary": "...",
  "entities_found": []
}

---

GET /api/history

Returns previously stored analyses.

---

POST /sanitize

Sanitizes raw text and returns findings.

---

Sensitive Data Detection (Current)

• Email addresses
• Phone numbers
• API keys
• Card-like patterns

---

Local Setup

1. Clone Repository

git clone https://github.com/hardik-bhalekar/SecureSum.git
cd SecureSum

---

2. Frontend Setup

cd client
npm install
npm run dev

Runs on:

http://localhost:3000

---

3. Backend Setup

cd server
npm install
npm run dev

Runs on:

http://localhost:5000

---

4. ML Service Setup

cd ml-service
python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt
uvicorn app:app --reload

Runs on:

http://localhost:8000

---

Environment Variables

Create server/.env

PORT=5000
MONGODB_URI=your_connection_string
ML_SERVICE_URL=http://localhost:8000

---

Folder Structure

SecureSum/
├── client/       # Next.js frontend
├── server/       # Express backend
├── ml-service/   # FastAPI service
├── README.md
└── .gitignore

---

Current Status

Completed

• Functional frontend
• Working backend APIs
• Sanitization microservice
• MongoDB persistence
• History retrieval
• End-to-end workflow

In Progress

• Premium UI redesign

Planned

• Real LLM summaries
• Search history
• Delete entries
• Authentication
• Audio transcription
• Deployment

---

Security Philosophy

SecureSum follows a simple principle:

sanitize first, summarize second

This reduces accidental exposure of sensitive information to downstream AI systems.

---

Portfolio Value

This project demonstrates:

• Full-stack engineering
• API integration
• Microservices architecture
• Security-first product thinking
• Database design
• Real-world SaaS workflow design

---

Author

Hardik Bhalekar

GitHub: https://github.com/hardik-bhalekar