Insights on AI, sports prediction, and building intelligent products
A technical look inside the architecture powering Scorient — from the Dixon-Coles parameter estimation pipeline to the real-time data ingestion layer. We explore the decisions that led us to reject standard Poisson models in favour of a corrected bivariate framework, and what that means for low-scoring match prediction accuracy.
The standard Poisson model is elegant but flawed — it systematically underestimates 0-0 and 1-1 draws. Dixon and Coles' 1997 correction changed everything for football statisticians. We ran backtests across 3 seasons of Premier League, La Liga, and Bundesliga data to quantify exactly how much it matters.
Goals are noisy. Expected Goals cut through the noise. A team can concede 3 goals from 0.4 xGA in one match and appear terrible — or score 0 goals from 2.8 xG and appear worse. We explain why Scorient layers xG onto its Dixon-Coles outputs and how this changes team strength assessments across a 38-game season.
Today we're launching Zoofy+ — an AI-powered pet behavior coach that analyzes your pet from a single photo. Breed detection, health markers, behavior signals, and personalized coaching. Available on iOS and Android for dogs, cats, rabbits, birds, and hamsters. Here's the full story behind the product.
Training a single model to accurately identify breeds across five distinct animal species — dogs, cats, rabbits, birds, and hamsters — is a fundamentally different challenge than single-species detection. We share what we learned: dataset construction, class imbalance, and how Gemini AI changed our architecture decisions entirely.
At AMIRTECH.AI, we don't pick a single AI provider and stop there. Scorient relies on custom statistical models and structured data pipelines. Zoofy+ uses Gemini's vision capabilities for photo analysis and Claude for coaching content generation. This is why we use both — and how they complement each other across our product portfolio.
2026 is the year AI stopped being a feature and became the foundation. On-device models, multi-modal analysis, real-time personalization — the gap between AI-native apps and traditional apps has never been wider. We survey the landscape and share where AMIRTECH.AI is placing its bets for the next 18 months.
What does it take to go from "what if AI could understand your pet from a photo?" to a live App Store product in under 6 months? We walk through the full journey — the initial concept, the tech stack decisions (Flutter, Supabase, Gemini, RevenueCat), the pivots, and the final push to launch. Honest, technical, and unfiltered.
Full articles are in progress. Subscribe to our newsletter for publication updates.
Stay Updated