title	Fraud Sentinel AI
emoji	🛡️
colorFrom	red
colorTo	blue
sdk	streamlit
sdk_version	1.43.0
app_file	app.py
pinned	false

🛡️ FraudSentinel AI: Agentic Scam Call Detection

Next-Generation Voice Fraud Analysis & Threat Intelligence

FraudSentinel AI is a state-of-the-art, multi-layered voice analysis platform engineered to definitively detect and intercept sophisticated phone scams. By merging ultra-fast phonetic transcription with deep semantic heuristics and Agentic LLM reasoning, the system generates real-time, explainable threat assessments for any audio interaction.

---

✨ Live Demonstration

🚀 Experience the live, unified dashboard in action:
👉 Launch FraudSentinel AI on Hugging Face

---

🧠 The 6-Layer Detection Architecture

Our system does not rely on simple word matching. It utilizes a synchronized multi-agent heuristic pipeline to analyze the psychological intent and semantic risk of a conversation:

1. 🎙️ Speech-to-Text Pipeline: Utilizes faster-whisper for highly optimized, offline transcription.
2. 📖 Semantic Engine: Deploys sentence-transformers to map conversational flow against high-risk scam embeddings.
3. ⚖️ Rule-Based Heuristics: Instantly flags critical trigger phrases (e.g., "gift card", "arrest warrant", "remote access").
4. 🎭 Social Engineering Analyzer: Detects psychological manipulation tactics like synthetic urgency, authority impersonation, and forced isolation.
5. 🤖 LLM Agentic Reasoning: Connects to via API OpenRouter to deploy advanced reasoning algorithms, contextualizing the conversation to eliminate false positives.
6. ⚠️ Final Risk Orchestrator: Fuses the sub-agent scores into a unified, actionable Threat Level (Safe, Suspicious, High Risk, Critical).

---

🛠️ System Workflow

graph TD
    A[Raw Audio / Live Call] -->|Whisper Transcription| B(Transcribed Text)
    B --> C{Agentic Orchestrator}
    C -->|Transformer Embeddings| D[Semantic Detector]
    C -->|Pattern Matching| E[Rule-Based Engine]
    C -->|Tactic Modeling| F[Social Engineering Detector]
    C -->|Contextual Reasoning| G[LLM API Engine]
    D --> H[Risk Fusion Engine]
    E --> H
    F --> H
    G --> H
    H --> I((Final Threat Report))

Loading

---

💻 Local Installation & Usage

You can easily run the FraudSentinel engine locally on your machine for maximum privacy and execution speed.

Prerequisites

• Python 3.9+
• FFmpeg (Required for audio processing)

Quick Start

1. Clone the highly secure repository:

   git clone https://github.com/coolss21/Agentic_AI_Scam_Call_Detection.git
   cd Agentic_AI_Scam_Call_Detection

2. Install the unified dependencies:

   pip install -r requirements.txt

3. Configure the AI Environment: Create a .env file in the root directory and add your OpenRouter key:

   OPEN_ROUTER_API_KEY=your_api_key_here

4. Launch the Core Interface:

   streamlit run app.py

---

🎨 UI/UX Excellence

The interface is wrapped in a breathtaking Cyber-Security styled glassmorphism aesthetic. The custom CSS automatically adapts to the analysis logic, deploying cinematic micro-animations, glowing risk-level indicators, and an ultra-modern dark mode palette guaranteed to impress.

🤝 Contribution & Maintenance

This repository is architected carefully for academic and commercial cyber-security evaluations. All pull requests are welcome.

Defending the vulnerable, one conversation at a time.