README.md

metadata

title: Janus ZeroTrust Sentinel
emoji: 🛡️
colorFrom: indigo
colorTo: blue
sdk: docker
app_file: app.py
pinned: true

🛡️ Janus: The Multimodal Intelligence Sentinel

Janus is a state-of-the-art, self-evolving Multimodal Intelligence System designed for high-fidelity financial research and proactive ZeroTrust threat interception. It operates as a cognitive agentic swarm that doesn't just process data—it dreams, reflects, and detects deception in real-time.

---

🧠 Core Intelligence: Cross-Modal Emotion Conflict Detection

"What you say ≠ how you feel."

One of Janus's flagship forensic capabilities is identifying Emotional Dissonance. Humans often rely on tone to detect sarcasm or deception; Janus automates this by measuring the divergence between spoken tone and transcript meaning.

🔬 The Dissonance Engine

• Audio Embedding: High-fidelity prosody analysis using wav2vec2 (ehcalabres/wav2vec2-lg-xlsr-en-speech-emotion-recognition).
• Text Embedding: Sentiment extraction using DistilBERT (bhadresh-savani/distilbert-base-uncased-emotion).
• Fusion Logic: Computes a Cosine Divergence Score between the two probability distributions.
• Dataset Calibration: Continuous self-refinement against the CMU-MOSEI and CREMA-D datasets to identify subtle sarcasm, irony, and emotional masking.

---

🌊 Cognitive Signal Flow

Every signal entering the Janus cluster undergoes a multi-stage cognitive transformation:

graph TD
    A[User Input / Multimodal Signal] --> B[InputProcessor]
    B --> C{SmartRouter 4D}
    
    C -- "Task: Finance" --> D[Domain Pack: AlphaVantage/NewsAPI]
    C -- "Task: Forensic" --> E[MMSA Dissonance Engine]
    C -- "Task: Research" --> F[Deep-Web Crawler / Jina Reader]
    
    D & E & F --> G[Knowledge Evolution Layer]
    G --> H[Cognitive Synthesis & Reflection]
    H --> I[Final Intelligence Brief]
    
    subgraph "The Reflection Loop"
        H --> J[Dream Engine: Speculative Scenarios]
        J --> K[Self-Reflection: Bias/Hallucination Check]
        K --> H
    end

---

🛡️ ZeroTrust & Forensic Guardian

Janus is built to intercept "Scam Journeys" before they manifest:

• Active Interception: Autonomous signal squashing for confirmed malicious trajectories.
• Relational Threat Mapping: Tracks scattered entities (phone numbers, URLs, crypto addresses) in the persistent ScamGraph.
• Forensic Safety Gateway: A dedicated portal for deep-probing SMS, chat logs, and malicious files using OCR and multimodal analysis.

📈 Financial & Adaptive Intelligence

• Autonomous Curiosity: Continuous background scanning of global markets and sentiment shifts via the Curiosity Engine.
• Mirofish Simulation: Predictive scenario modeling (Monte Carlo) for market volatility and opportunity forecasting.
• Knowledge Ingestion: Automatic distillation of web evidence into a persistent, freshness-aware knowledge base.

---

🚀 Deployment & Tech Stack

Tech Stack

• Backend: FastAPI, PyTorch, Transformers (wav2vec2, BERT), MediaPipe, librosa.
• Frontend: Next.js 14, Tailwind CSS, Lucide React, Framer Motion.
• Data: ScamGraph (Relational), Mirofish (Simulation), Jina (Web Reader).

Setup & Run

Janus is fully containerized. To run the complete cluster locally:

# 1. Clone the repository
git clone https://github.com/DevodG/Janus.git
cd Janus

# 2. Run the unified development environment
./run-dev.sh

Environment Secrets

To enable full cognitive capability, configure the following in .env:

• HUGGINGFACE_API_KEY: Core reasoning models.
• TAVILY_API_KEY: Deep-web research access.
• NEWS_API_KEY: Real-time financial signals.
• ALPHAVANTAGE_API_KEY: Market data integration.

---

Janus is an Autonomous Learning System. Every interaction refines its internal "Skills" and "Prompt Weights," making it more precise with every case it resolves.

Janus Adapts. Janus Dreams. Janus Protects.