Project Darpan

Darpan (दर्पण), meaning “mirror,” is an analysis engine that reflects the true nature of online content, exposing toxicity, misinformation, and hate that weaken healthy digital discourse.

Comment

Project Darpan

Reflecting the Truth Behind Online Content

December 24, 2025

About the Project

Darpan, meaning mirror in Sanskrit, is an AI-powered browser extension designed to reflect the true nature of online discourse. It analyzes social media content in real time to detect toxicity, propaganda, and coordinated misinformation campaigns targeting India.

The project was developed for Code Spring 2025 with the goal of empowering users through transparency and contextual awareness, rather than opaque moderation.

Problem Statement

Digital platforms such as X (Twitter), YouTube, Telegram, and Facebook are frequently misused to spread:

  • Anti-India propaganda
  • Hate speech and communal incitement
  • Coordinated misinformation campaigns

Existing moderation systems are often reactive, centralized, and non-transparent.
There is a strong need for a real-time, user-side, explainable AI solution.

Inspiration

The project was inspired by:

  • Lack of India-specific context in generic toxicity detectors
  • Black-box content moderation systems
  • The growing importance of digital literacy and informed consumption

Darpan focuses not only on flagging content, but also on clearly explaining why it is flagged.

Approach

Project Darpan follows a hybrid multi-layered pipeline:

1. Local Rule-Based Analysis

  • Keyword and phrase detection for hate speech, disinformation, and separatist rhetoric
  • Identification of propaganda tactics such as excessive capitalization and emotional triggers

2. Threat Scoring System

Each detected pattern contributes to a cumulative threat score:

Threat Score = Σ (wᵢ × fᵢ)

Where:

  • fᵢ = detected feature
  • wᵢ = assigned severity weight

3. AI-Powered Contextual Analysis

For deeper understanding, the system integrates the Hugging Face Inference API using the model:

cardiffnlp/twitter-roberta-base-sentiment-latest

Tech Stack

Programming Languages

  • JavaScript – Analysis engine, API calls, DOM manipulation
  • HTML – Popup interface structure
  • CSS – User interface styling

Frameworks and Libraries

  • Web Extensions API
  • Tailwind CSS

APIs

  • Hugging Face Inference API

Innovation and Uniqueness

  • India-specific toxic pattern database
  • Category-based analysis instead of a single toxicity score
  • Transparent explanation of detected issues
  • Real-time, on-screen feedback while browsing

Challenges Faced

  • Difficulty in detecting sarcasm and satire
  • Dependency on external APIs for large-scale deployment
  • Frequent UI and DOM changes on social platforms

Mitigation Strategies

  • Hybrid AI approach (local + ML)
  • Local fallback analysis engine
  • Modular and adaptable extraction logic

Impact and Benefits

  • Empowers users to identify propaganda
  • Enhances digital and media literacy
  • Reduces exposure to harmful and manipulative content

Key benefits include:

  • Instant verdicts
  • Numerical threat scores
  • Detailed key findings

Key Learnings

  • Importance of explainable AI over opaque moderation
  • Effectiveness of browser extensions for user-side AI
  • Transparency significantly improves user trust

Project Links

References

  • Hugging Face Inference API Documentation
  • Political Hate Speech Detection in Indian Elections
  • Disinformation and Psychological Warfare Reports

Note: This project was built entirely from scratch and represents an original idea by me.

Built With

Share this project:

Updates