Aim

To provide a private, localized communication ecosystem for the NITK community that ensures student anonymity while maintaining safety through internal network restrictions and automated moderation
 

Introduction

Online student forums often face issues with privacy and external interference. Ripple addresses this by creating a walled garden, a platform accessible only via NITK-NET. By requiring verified NITK email addresses for login, the app ensures that every anonymous user is a legitimate member of the student body, fostering a safe space for campus-wide polls, department discussions, and peer-to-peer networking.
 

Literature survey and technologies used

The project leverages a modern web development stack combined with automated natural language processing (NLP) and localized network protocols to ensure a secure campus-only ecosystem.

 

Core Tech Stack:

Frontend: React.js, Vite, and Tailwind CSS for a high-performance, responsive UI.

Backend API: Node.js and Express.js to manage RESTful endpoints and core business logic.

Real-time Communication: Socket.io for persistent, WebSocket-based bi-directional event transmission.

Database & Caching: Supabase (PostgreSQL) for persistent data storage and Upstash Redis for low-latency message enqueuing.

AI Moderation Worker: A dedicated Python microservice utilizing the Hugging Face Transformers library (unitary/toxic-bert) for local, high-speed NLP toxicity analysis.

 

Technical Feature Implementation

Network-Level Access Control (Intranet Restricted): The system utilizes server-side IP filtering to validate that incoming socket connections originate from NITK’s authorized CIDR blocks (NITK-NET).

Identity & Verification Layer: Integration of NITK-specific SMTP verification to restrict account creation exclusively to @nitk.edu.in domain, ensuring a verified user base.

Automated Room Provisioning: A logic layer that parses user metadata to auto-assign clients to specific department and year-wise PostgreSQL schemas/tables upon first login.

Anonymous Identity Mapping: A secure hashing mechanism that maps verified NITK emails to unique, session-based anonymous identifiers (e.g., Anon-4D0D), decoupling real identity from the message layer.

Real-time Content Redaction: A decoupled AI worker that monitors the Redis moderation_queue. If a message exceeds the toxicity threshold, the worker triggers a database update and a WebSocket broadcast to replace the offensive string across all active client states.

Methodology

Ripple operates on a distributed architecture dividing client-facing real-time communication and backend computational tasks:

1. Network Validation & Intranet Handshake:

• Before authentication, the system performs a handshake to verify the client's connection originates from the NITK-NET.

• Access is restricted via server-side IP filtering to ensure the platform remains a localized "walled garden" for the campus community.

2. Verified Authentication & Anonymization:

• Users must authenticate using @nitk.edu.in to ensure only legitimate students can access the platform.

• Upon successful login, the system decouples the user's real identity from their chat persona by assigning a unique, session-based anonymous identifier (e.g., Anon-FC51).

3. Real-Time Communication & Queueing:

• Once connected to the Socket.io endpoint, users can exchange messages in real-time or participate in anonymous campus-wide chat groups.

• When a message is transmitted, the Node.js server simultaneously broadcasts it to the chat room and enqueues the content into a Redis list (moderation_queue) for asynchronous processing.

4. AI Inference Workflow:

• A separate, localized Python worker.py continuously polls the Redis queue to ensure data remains within the campus network.

• It performs deep NLP analysis on the dequeued text using the unitary/toxic-bert model, which classifies content across various toxicity dimensions.

5. Retraction Protocol & Governance:

• If the toxicity score exceeds the safety threshold, the Python worker flags the message in the Supabase database.

• The worker then publishes a retraction event back to Node.js via Redis Pub/Sub, triggering an immediate WebSocket emission to all clients.

• The offensive text is dynamically replaced with a community guideline violation notice on the client interface, maintaining a safe environment without interrupting the real-time flow.

 Results

Ripple successfully operates as a low-latency communication hub. Testing shows that the restricted login and network protocols effectively wall off the community, while the AI moderation processes messages within milliseconds, ensuring that the anonymity of the platform is not weaponized for harassment.

Future Scope

Ripple is a secure, localized messaging ecosystem engineered exclusively for the NITK Surathkal student body. Operating strictly within the institute’s internal Wi-Fi and LAN infrastructure, the platform ensures a "walled garden" environment where access is restricted to users with verified NITK email addresses.

The application facilitates candid, real-time communication through an anonymous identity mapping system, allowing students to interact in student-made groups with direct chats. To maintain community standards, Ripple integrates a distributed AI moderation pipeline; a dedicated Python worker utilizes the unitary/toxic-bert model to perform deep NLP analysis, automatically redacting toxic content via a Redis-backed retraction protocol without compromising the user's anonymity or the platform's low-latency performance.

Future enhancements may include:

• Optimizing the AI inference to run efficiently on serverless cloud environments (like Hugging Face Inference API) for reduced cloud hosting costs.

• End-to-end encryption for message privacy prior to text analysis.

• Multimedia support (images and audio) along with computer vision-based moderation.
   

References

GitHub Repository: https://github.com/20Varun23/Ripple

Transformers model: https://huggingface.co/unitary/toxic-bert
 

Mentors mentees details

Mentors:        

VARUN ACHARYA KV

DINESH MANIDEEP PALUVADI

Mentees:

SHASHANK S

AMAN NAGPAL