How lordzedith/anomaly-detection-platform works
Project Type and Purpose
• Type: Full-stack web application with a separate ML pipeline.
File Structure
• Top-level directories:
• anomaly_models: Python ML training + prediction code, data, results
• server: Node.js/TypeScript backend (Express style)
• src: React/TypeScript frontend (Vite)
• Key files:
• anomaly_models/run_prediction.py & ai_chat.py
• server/index.ts, server/routes/*.ts, server/config/database.ts
• src/App.tsx, src/api/api.ts, components, hooks
• Config/build files:
• README.md, .env, package.json, tsconfig*, tailwind.config.js, Vite configs, requirements.txt
Extract Insights from the README
• ML models: 225 total, 117 active, 9 algorithms, 25 strategies
• Features: real-time detection, AI Robo Advisor (OpenAI), visualization, news integration
• Tech stack: Python 3.12 (scikit-learn, NumPy, joblib), Node.js/TypeScript, React/Vite, Tailwind CSS, OpenAI API, relational DB
Main Components
a. Users & Clients
• Web browser / UI (React + Vite dev server or static hosted)
b. Frontend Application
• React App
• API client module (src/api/api.ts)
• UI components (charts, dashboard, strategy manager)
c. Backend API Server
• Node.js/Express (server/index.ts)
• Middleware (auth, installation)
• Routes/controllers (anomaly, auth, strategies, install)
• Database connector & migrations (server/config)
d. Database
• Relational DB (PostgreSQL or MySQL)
• Tables: users, strategies, installations, perhaps model metadata
e. ML Pipeline (Python)
• Training module (anomaly_models/prepared_data → anomaly_models/results)
• Prediction service (run_prediction.py)
• AI chat module (ai_chat.py calling OpenAI API)
• Model artifact storage (joblib / pkl files under results/)
f. External Services
• OpenAI API (AI Robo Advisor)
• Financial data feeds (stock prices, news sources – implied)
Relationships & Data Flow
• User interacts with React UI → sends HTTP requests to Backend API
• Backend API authenticates user → reads/writes from Database
• On anomaly detection request:
• Backend invokes Prediction Service (either via REST call, CLI spawn, or microservice endpoint)
• Prediction Service loads model artifacts → returns anomaly results
• Backend stores/updates results metadata in Database → returns response to frontend
• For AI Robo Advisor:
• Backend or Prediction Service calls ai_chat.py → communicates with OpenAI API → returns advisory
• Offline Model Training:
• Data engineers run anomaly_models training scripts → ingest prepared_data → output model artifacts in results/
• Results are versioned on disk or in artifact repository
Architectural Patterns
• Client-Server (React front-end vs Node.js API)
• Microservice-like separation of ML pipeline (Python) from API server
• Layered architecture: presentation → API business logic → data persistence → model/prediction layer
• Use of environment-based configuration (.env for ports, API keys)
• Infrastructure as code: migrations for DB schema (SQL scripts)
Technologies & Frameworks to Annotate
• Frontend: React, Vite, TypeScript, Tailwind CSS
• Backend: Node.js, Express (TypeScript), ts-node, Knex or TypeORM (implied), JWT auth
• ML Pipeline: Python 3.12, scikit-learn, NumPy, joblib/pickle, pandas
• External APIs: OpenAI (ChatGPT), market data providers
• Database: PostgreSQL (or similar)
Final Checklist
• Components: User, Frontend, Backend API, ML Pipeline, Database, External APIs
• Flows: HTTP requests, DB queries, model loading, API calls to OpenAI
• Patterns: client-server, microservice, layered architecture