Case Study · 2026
E-commerce Development
Transform your retail business with our comprehensive e-commerce solutions. We build secure, scalable, and user-friendly online stores that provide an exceptional shopping experience. From platform selection (Shopify, WooCommerce, Magento) to custom payment gateway integration, we handle everything to get you selling online effectively.
Context
Oigetit filters fake news with an AI scoring engine. I sat in the loop — validating predictions, hunting edge cases, and explaining the machine to the humans it serves.
The Problem
An accurate model nobody understands is an untrusted model. The work was double-sided: make the AI more right, and make its rightness legible.
Process
Validate at the edges
Systematic pattern recognition across misclassifications — sarcasm, partial truths, source laundering — fed back to strengthen the pipeline.
Translate the engine
Rewrote how the verification engine explains itself: simplified, user-facing language for why an article scores the way it does.
Design Decisions
Explanations in the reader's language
'Confidence: 0.82' persuades no one; 'multiple independent sources confirm the core claim' does.
Reflection
“AI products are trust products. The interface between a model and a person is exactly as strong as its explanation.”