Skip to content

README.md

Latest commit

 

History

History
62 lines (36 loc) · 3.04 KB

README.md

File metadata and controls

62 lines (36 loc) · 3.04 KB
Raw

Agentic AI Automated Trading System

Project Overview

MSC Finance Project — An end-to-end agentic AI pipeline for automated trading signal generation across 11 financial assets using machine learning, technical analysis, and AWS cloud infrastructure.

The entire pipeline is deployed on AWS, operating daily via serverless triggers, with a real-time monitoring dashboard hosted on an EC2 instance.

System Architecture & Workflow

The system is divided into six distinct layers, separating the heavy data science processing from the lightweight daily execution and user interface.

  • Agent 1 (Market Analysis): Calculates technical indicators (RSI, MACD, SMA crossovers) and labels data based on 5-day forward returns.

  • Agent 2 (Information Retrieval): Fetches daily global news headlines and extracts multi-class sentiment scores using NLP.

  • Agent 3 (Decision Engine): A Random Forest Classifier (100 Trees, 13 Features) that outputs BUY/SELL/HOLD signals with confidence probabilities.

  • Agent 4 (Risk Management): The "Gatekeeper." Enforces a 55% minimum confidence threshold, stop-loss limits, and portfolio sizing rules before approving a trade.

    Duagram 1 drawio
Diagram 2 drawio diagram 3 drawio

Technology Stack

  • Cloud Infrastructure (AWS): * EventBridge: Daily cron schedule trigger (8:00 AM Local).
    • Lambda: Executes the daily trading pipeline.
    • S3: Stores historical raw data, trained .pkl models, and daily signals.json outputs.
    • RDS (MySQL 8.4): Maintains an immutable audit trail (trade_log) of all AI decisions.
    • EC2: Hosts the persistent 24/7 web dashboard.
  • Machine Learning: scikit-learn (Random Forest, StandardScaler)
  • Data Pipelines: yfinance (OHLCV data), NewsAPI (Sentiment), pandas
  • Frontend UI: Streamlit

Project Structure

├── data/

│ ├── featured/ # Processed data with technical indicators

│ ├── plots/ # Generated EDA and performance charts

│ ├── raw/ # Raw OHLCV market data

│ ├── backtest_summary.csv # Historical backtesting results

│ └── sentiment_scores.csv # NLP sentiment analysis outputs

├── notebooks/ # Jupyter notebooks for experimentation

├── package/ # AWS Lambda deployment packages

├── src/ # Python scripts for the 4 ML Agents

├── .gitignore # Hidden files and keys

├── app.py # Streamlit Dashboard UI

├── requirements.txt # Project dependencies

└── README.md # Project documentation