Apppublic

im-amrith/nerve

sourceHugging Faceupdated 8mo agoView on Hugging Face

0likes

App README

Agentic Brokerage OS

An intelligence-first brokerage layer that governs trading through autonomous AI agents.

🎯 Core Philosophy

Problem: Brittle execution-only models lead to retail losses through emotional trading and lack of behavioral guardrails.

Solution: Self-healing adaptive automation using Vision Language Models and multi-agent orchestration.

� Interactive Dashboard (Recommended for Judges)

We've built a beautiful web-based dashboard to showcase all features interactively!

One-Click Launch (Windows):

bash

run_dashboard.bat

One-Click Launch (Linux/Mac):

bash

chmod +x run_dashboard.sh
./run_dashboard.sh

Manual Launch:

bash

# Install Streamlit
pip install streamlit>=1.30.0

# Launch dashboard
streamlit run dashboard.py

Then open your browser to: http://localhost:8501

The dashboard provides:

—✅ Interactive demos of all 4 core features
—✅ Real-time visualization of AI agents
—✅ Live code generation and execution
—✅ Beautiful UI for judges to evaluate
—✅ One-click feature switching

�🏗️ Architecture

Perception (CV + Groq VLM) → Reasoning (LangGraph) → Action (Trading APIs)

Core Components

—Orchestrator: LangGraph-based multi-agent coordination
—Perception Engine: Computer Vision + Groq for UI understanding (no DOM selectors)
—Reasoning Engine: LLM-based planning and decision-making
—Pre-Trade Sentinel: Sub-50ms safety checks and kill switch
—Strategy Engine: Natural language to executable trading algorithms
—RAG Journaling: Context-aware trade autopsy system

🚀 Quick Start

Prerequisites

bash

Python 3.10+
pip install -r requirements.txt

Environment Setup

Create a .env file:

env

GROQ_API_KEY=your_groq_api_key
PINECONE_API_KEY=your_pinecone_key
PINECONE_ENV=your_environment

Run the System

bash

# Start the orchestrator
python src/main.py

# Run demos
python demos/ui_adaptation_demo.py
python demos/strategy_generation_demo.py

📁 Project Structure

zerodha/
├── src/
│   ├── agents/              # Specialized agents
│   │   ├── perception_agent.py
│   │   ├── reasoning_agent.py
│   │   ├── sentinel_agent.py
│   │   └── strategy_agent.py
│   ├── core/                # Core systems
│   │   ├── orchestrator.py  # LangGraph workflow
│   │   ├── perception.py    # CV + VLM perception
│   │   └── state.py         # Agent state definitions
│   ├── engines/             # Business logic
│   │   ├── pre_trade_sentinel.py
│   │   ├── strategy_engine.py
│   │   └── rag_journal.py
│   ├── utils/               # Utilities
│   │   ├── vision.py
│   │   ├── memory.py
│   │   └── validators.py
│   └── main.py              # Entry point
├── demos/                   # Demonstrations
├── tests/                   # Test suite
├── config/                  # Configuration files
└── docs/                    # Documentation

🎪 Demos

1. UI Adaptation Demo ("Sabotage Test")

Demonstrates self-healing when UI elements are moved/modified.

bash

python demos/ui_adaptation_demo.py

2. Strategy Generation

Convert natural language to backtested trading algorithms.

bash

python demos/strategy_generation_demo.py

3. Pre-Trade Sentinel

Live demonstration of the kill switch blocking risky trades.

bash

python demos/sentinel_demo.py

🔐 Security Features

—Zero-Knowledge ML: Verifiable inference without data exposure
—User Constitution: Personalized trading rules and guardrails
—Kill Switch: Hard blocks on rule violations (<50ms latency)

📊 Key Metrics

—Perception Latency: <200ms for screen-to-JSON mapping
—Sentinel Response: <50ms for trade safety checks
—UI Adaptation: Real-time self-healing without code changes

🛠️ Technology Stack

—VLM: Groq (llama-3.2-90b-vision-preview)
—Agent Framework: LangGraph
—Vector DB: Pinecone
—Computer Vision: OpenCV + Pillow
—API Layer: FastAPI (future)

📝 License

MIT

🤝 Contributing

See CONTRIBUTING.md