AIFORUS
AI-news intelligence — built solo, shipped to production.
The problem
Non-technical teams needed to track global AI developments without reading hundreds of sources every day.
What it does
A backend continuously discovers AI-related news URLs from global media (the "clt" service), then fetches and normalizes the content ("scr" service) on a scheduler; an ML layer scores articles for AI-relevance. A React + Vite dashboard with Leaflet maps and Recharts turns the feed into something a non-technical user can scan. Live in production at aicerti.co.kr, with a stripped-down public demo (FastAPI + React, sample data) open-sourced.
Stack
- Python
- FastAPI
- React
- Vite
- Leaflet
- Recharts
- Docker
- nginx
My role
- Solo engineer (backend + frontend)
- Pipeline & scheduler architecture
- ML integration
- Docker / nginx deploy
Highlights
Shipped to production
Designed and built the whole system solo — collection, normalization, ML scoring, API, dashboard and deploy — running in production at aicerti.co.kr.
Two-service pipeline
Decoupled URL discovery (clt) from content fetch + normalization (scr); each runs alone or together via an operator layer + scheduler, for stable long-term collection.
Zero-shot relevance
Scores incoming articles for AI-relevance with no per-label training data, so new topics need no retraining.
Built for non-experts
Map and trend views turn a noisy global feed into something readable at a glance.
Screens
The production system is company IP. The linked repo + live demo are a public, stripped-down version (sample data) of the same architecture.