Neutrino

This project is an AI-driven quant trading platform that allows users to create, backtest, and deploy buy/sell trading strategies using plain language commands. It bridges the gap between human intuition and algorithmic execution by combining natural language understanding, financial data analysis, and multi-agent intelligence.

How It Works

User Interface Layer Users describe their trading strategies in simple English (e.g., “Buy BTC when RSI < 30 and sell when RSI > 70”).

The system translates this into an executable quant algorithm.

Users can backtest strategies on historical market data before approval.

Once approved, strategies move to the live trading environment.

Algorithm Engine

Integrates real-time market data and price feeds from multiple exchanges.

Continuously runs approved algorithms to generate buy/sell signals based on current market conditions.

Works in sync with the intelligence layer for improved signal accuracy.

Intelligence Layer (Agentic System) The Supervisor Agent coordinates specialized AI agents that enhance decision-making by inter-agent coordination.

Technical Agent — performs technical analysis using indicators and price action.

Sentiment Agent — processes market sentiment from news, social media, and other sources.

Web Search Agent — retrieves real-time external information influencing market moves.

Trading Agent — combines insights to recommend or validate trading actions. The Supervisor Agent evaluates all inputs and decides whether to approve or reject each signal before execution.

Execution Layer

The Execution Engine places approved orders on supported platforms such as Hyperliquid, other DEXs, and CEX integrations.

The Monitoring Engine tracks all executed trades, updating system performance and feeding data back for strategy optimization.

Core Value

This system lets anyone—from novice traders to professionals—design sophisticated quant strategies without writing code. It combines AI reasoning, real-time data, and multi-exchange automation into one unified platform capable of evolving with the market.

This project is an advanced algorithmic trading system that combines modern web technologies, agentic AI architecture, and real-time financial market integration. Here's the technical breakdown: Frontend Architecture (Next.js + Vercel) The user interface is built with Next.js and deployed on Vercel for optimal performance and scalability. Key features include:

Avail Nexus SDK Integration: This is particularly noteworthy - we integrated Avail's bridging solution directly into the UI, allowing users to deposit tokens to Hyperliquid in a single, seamless flow. This eliminates the friction of multi-step bridging processes. WebSocket Communication: Real-time bidirectional communication with the agentic backend enables live updates of trading signals, agent reasoning, and market analysis. Privy Wallet Integration: Handles secure wallet connections and authentication, ensuring users maintain custody of their funds.

Execution Engine (Node.js + MongoDB) The execution layer acts as the critical middleware between the AI agents and the trading infrastructure:

API Wallet Management: A particularly secure implementation where API wallets can only place trades but cannot withdraw funds from user accounts - users maintain full custody. Hyperliquid Client Integration: Direct integration with Hyperliquid DEX for order execution on both spot and perpetual markets. JWT Authentication: Access tokens from Privy ensure secure, stateless authentication across the platform. MongoDB: Stores user profiles, trading strategies, custom agent configurations, and historical performance data.

Agentic AI Backend (Python + LangGraph) This is where the magic happens - a sophisticated multi-agent system built with: Core Framework

LangGraph: Orchestrates complex agent-to-agent communication flows and maintains conversational state across the trading decision pipeline. LangChain: Provides the foundation for agent tooling and LLM interactions. Model Context Protocol (MCP): Enables standardized communication between agents and their respective tools - this is a particularly clean architectural choice that makes the system modular and extensible. FastAPI WebSocket: Provides real-time streaming of AI reasoning and decisions to the frontend. AWS EC2 Deployment: Ensures 24/7 availability for continuous market monitoring.

The Supervisor Agent Pattern The Supervisor Agent coordinates four specialized worker agents, each with distinct capabilities:

Finance Agent (Technical Analysis)

Integrated with taapi.io API providing access to 200+ technical indicators Supports analysis across multiple timeframes and intervals Can identify patterns like moving average crossovers, RSI divergences, MACD signals, Fibonacci retracements, and much more This breadth of indicators makes the system highly adaptable to different market conditions

Sentiment Agent (Real-time Market Sentiment)

Uses Google News API to fetch the latest financial news Employs a custom ML sentiment analysis model to gauge market sentiment in real-time Particularly valuable for catching sudden market shifts driven by news events

Trading Agent (Risk Management & Execution)

Implements a custom algorithm for stop-loss and take-profit calculation Analyzes support and resistance levels in real-time to determine optimal entry/exit points Performs risk-to-reward analysis before approving trades This is the final gatekeeper ensuring only favorable trades are executed

Web Search Agent (General Information Retrieval)

Handles ad-hoc information needs that fall outside the specialized domains Can research specific assets, market conditions, or trading concepts on demand