Runes ML Forecasting Pipeline
A multi-container, full-stack ML platform that collects, forecasts, and visualizes cryptocurrency ("Runes") market data — a custom-trained LSTM served through Dockerized microservices with a real-time React dashboard.
- Python
- Flask
- LSTM / TensorFlow
- Node.js
- React
- PostgreSQL
- MongoDB
- Redis
- Docker
- Prometheus / Grafana
Overview
A modular, multi-container application that collects, processes, forecasts, and visualizes market data for the "Runes" cryptocurrency project. Each Docker container owns one role — data collection, forecasting, serving, caching, monitoring, and the UI — so the system is scalable and independently deployable.
Architecture
A data-collection service ingests rune metrics (price, volume, holders) from external APIs into PostgreSQL and MongoDB, while a companion service scrapes relevant Discord discussion for cross-referencing. A Flask forecasting service preprocesses the series and runs a custom-trained LSTM to produce 4-hour and 24-hour forecasts, with an on-demand retraining module that streams live epoch metrics to the dashboard.
A Node API server serves extensive time-series data to a React dashboard, backed by a Redis caching layer for fast retrieval. Every service exposes a /health endpoint surfaced as green/red status lights in the UI, and cAdvisor + Prometheus + Grafana provide real-time container and system metrics (CPU, memory, network I/O).