Samuel Gicho@samuel_gicho666
1
Events attended
1
Submissions made
6+ years of experience
About me
I’m a Software and AI Engineer with a strong passion for building intelligent, scalable, and user-focused applications. I specialize in Python, FastAPI, Django, React/Next.js, and cloud deployment, with hands-on experience designing and implementing AI-driven solutions, backend systems, and automation workflows. Through programs like the ALX Pro Backend Developer course and the Google Africa Developer Scholarship (Associate Cloud Engineer), I’ve strengthened my expertise in backend development, cloud computing, system architecture, and AI integration. I enjoy solving real-world problems by combining clean code, modern frameworks, and practical AI tools. I love collaborating with others, learning new technologies, and building products that deliver value. I am always exploring opportunities to apply AI responsibly and creatively, especially in contexts that improve access, efficiency, and user experience.
🤓 Latest Submissions
FraudShield AI Real-Time Fraud Detection
FraudShield AI tackles the $32+ billion annual fraud problem in financial services with an intelligent, real-time detection platform. The Problem Traditional rule-based systems have 80-90% false positive rates, frustrating customers and overwhelming analysts. Modern regulations (EU AI Act, GDPR) demand explainable AI decisions, not black-box models. Our Solution We built an ensemble machine learning system combining three complementary approaches: - Isolation Forest (30%) - Catches novel, unknown fraud patterns through unsupervised anomaly detection - XGBoost (50%) - High-accuracy supervised classification on known fraud types - Heuristic Rules (20%) - Domain-expert rules for regulatory compliance and interpretability Key Features - Real-time Analysis: Sub-100ms scoring per transaction - Explainable AI: Every decision includes natural language explanations ("Large $45,000 trade at 2:34 AM with velocity burst detected") - Risk Scoring: Continuous 0-100 scores instead of binary fraud/not-fraud - Human-in-the-Loop: Analysts can confirm or dismiss alerts, feeding back into model improvement - 15+ Fraud Signals: Velocity attacks, amount anomalies, off-hours trading, account takeover patterns, and more Technical Stack - Backend: FastAPI (Python) with scikit-learn and XGBoost - Frontend: Next.js 14 with real-time dashboard - Deployment: Vercel (frontend) + Render (backend) Impact Our calibrated system targets realistic fraud rates (1-5%), dramatically reducing false positives while maintaining high recall on actual fraud. The explainability layer ensures compliance with emerging AI regulations while building analyst trust. Live demo available, experience real-time fraud detection with synthetic trading data.
7 Feb 2026
