Update on Building in Public • 6-month challenge 🚀

Decided to go with this direction for the hero section of my first Framer template.

Lots of iterations ahead, but I'm excited to see where this goes.

More updates coming soon. 👨‍💻

This is living rent-free in my head.

🏆 I built a live AI-powered match predictor for the 2026 FIFA World Cup — and it updates itself every hour, automatically. 48 teams. 104 matches. The biggest sporting event on the planet. So I asked myself: can machine learning predict what no one can? 🤖 𝗧𝗛𝗘 𝗠𝗢𝗗𝗘𝗟 Not just another ELO-based predictor. This one combines: ⚡ Live ELO ratings — recalculated after every WC match 📈 Recent form — last 5 & 10 official matches ⚔️ Head-to-head history — including World Cup-only records 👥 Squad strength — EA FC 26 player ratings 💰 Market value — Transfermarkt squad valuations 🎯 Penalty shootout win rate — the "clutch factor" most models ignore 🔥 Knockout pressure — encoded as a binary feature ⚠️ Upset detection — flags weaker teams in better form Dual-model ensemble: XGBoost + Poisson Regression 22 engineered features across 9 categories Trained on 49,477 international matches (1872–2026) 🔧 𝗧𝗛𝗘 𝗥𝗘𝗔𝗟 𝗘𝗡𝗚𝗜𝗡𝗘𝗘𝗥𝗜𝗡𝗚 𝗖𝗛𝗔𝗟𝗟𝗘𝗡𝗚𝗘 Building the model wasn't the hard part — keeping it LIVE was. Results change daily, teams get eliminated mid-tournament, and shootouts aren't always recorded immediately. The solution: an update engine that runs every hour — fetching results, detecting eliminated teams, recalculating ELO, and handling shootouts with smart fallback logic. Fully automated. 📊 𝗧𝗛𝗘 𝗕𝗜𝗚𝗚𝗘𝗦𝗧 𝗧𝗛𝗜𝗡𝗚 𝗜 𝗟𝗘𝗔𝗥𝗡𝗘𝗗 Early versions suffered from "ELO Dominance" — assuming the stronger team almost always wins. Statistically wrong. The fix: convert raw ELO difference into a calibrated win probability using the standard logistic formula. Now a +300 ELO advantage means ~88% win probability, not 100%. Because football is unpredictable. That's what makes it beautiful. 🖥️ 𝗧𝗛𝗘 𝗗𝗔𝗦𝗛𝗕𝗢𝗔𝗥𝗗 A 6-page Streamlit app featuring: 🔮 Match prediction — Win/Draw/Loss probabilities 🔵 Group stage simulation — Monte Carlo standings 🏆 Championship probability per team 📄 Downloadable PDF report 🌙 Dark/Light mode 🇸🇦 Arabic / 🇬🇧 English support 🎬 Full walkthrough in the video below 👇 🔴 Live App → https://lnkd.in/e8BuSUh7 🐙 GitHub → https://lnkd.in/ej8w_cUq ⚠️ Built for educational & portfolio purposes. Predictions are probabilistic — not guarantees 😄 What feature would YOU add to improve upset prediction? Drop it below 👇 #MachineLearning #DataScience #Streamlit #SportsAnalytics #FIFA2026