EthioMedIntel
Medical Intelligence Platform
An end-to-end medical data intelligence platform that scrapes, transforms, and classifies Telegram health data using production-grade ETL pipelines and YOLOv8 computer vision.
System Architecture
Screenshots
Problem
Ethiopian medical professionals share critical health information through Telegram channels, but this data is unstructured, scattered, and inaccessible for systematic analysis. There was no pipeline to ingest, clean, model, and serve this data for decision-making.
Architecture & Pipeline
Built a full ELT pipeline: Telethon scrapes 650+ messages from medical Telegram channels, dbt transforms raw data through staging, intermediate, and mart layers, Dagster orchestrates the entire pipeline with scheduling and monitoring, and PostgreSQL star schema stores the dimensional model.
ML & Computer Vision
Integrated YOLOv8 for medical image classification, achieving 80%+ accuracy on health-related visual content. The model classifies images shared in Telegram channels to detect medical conditions and categorize content types.
API & Dashboard
Developed a FastAPI backend with 10+ RESTful endpoints serving cleaned, transformed data. Built a Next.js 16 dashboard with interactive visualizations, real-time data exploration, and comprehensive analytics views.
Impact & Results
Processed 650+ medical messages through a fully automated pipeline. Achieved 80%+ image classification accuracy. Delivered 10+ API endpoints with sub-200ms response times. Created a production-ready dashboard for medical data analysis.
Interested in this project?
Let's discuss the engineering decisions and system design.