Even though modern messaging apps like WhatsApp and Telegram protect message content using End-to-End Encryption (E2EE), they still leak **routing metadata — such as sender, recipient, timestamp, and message size.
Attackers or service providers can analyze this metadata to infer communication patterns, social graphs, or even who talks to whom.
Our project presents a working prototype that introduces adaptive privacy layers to obscure such metadata without sacrificing delivery performance.
It combines simulation, AI-driven attacker evaluation, and real-time reasoning using Pathway and LLM intelligence.
📊 Final Presentation Slides — PPT Link Here
- Project Overview
- Problem Background
- Existing E2EE Systems (WhatsApp, Telegram)
- Proposed Solution
- Privacy Layer Implementation
- LLM-Based Adversary (Pathway Integration)
- Results & Metrics
- Future Scope
| Category | Tools / Frameworks Used |
|---|---|
| Language & Simulation | Python (asyncio, pandas, numpy) |
| AI & Privacy Evaluation | OpenAI API (LLM), Pathway Framework (Real-Time RAG) |
| Visualization / Dashboard | Streamlit |
| Machine Learning (Attacker Heuristics) | TF-IDF Vectorization, Clustering, Correlation Analysis |
| Version Control & Config | Git, YAML, dotenv |
- Adaptive Privacy Layer — Adds batching, padding, dummy routing to break linkability patterns.
- LLM-Based Adversary Simulation — AI-generated attack heuristics for robust privacy evaluation.
- Pathway Integration — Real-time RAG index for live attack parameter updates using streaming data.
- Quantitative Metrics — Calculates linkability %, latency trade-offs, and visualization dashboards.
- Modular Design — Easily extendable to new privacy techniques and attacker models.
Team Name: Zyphers
- ROHIT S
- S ABISHEAK
- S AKILESH
- ROOPESH C