BI Scientist || AI/ML Engineer || Python Developerhttps://ikebude.github.io/https://ikebude.github.io/blog/can-100-dollars-become-4000/https://ikebude.github.io/blog/can-100-dollars-become-4000/Lolo's Ledger has a goal reality check: a Monte Carlo simulation that tells a trader, before they start, whether their target is possible at safe risk.Sat, 31 Oct 2026 00:00:00 GMThttps://ikebude.github.io/blog/credit-score-cutoff-is-a-business-decision/https://ikebude.github.io/blog/credit-score-cutoff-is-a-business-decision/LoanLens scores 55,518 real Prosper loans. The model ranks borrowers; the cutoff that decides who gets a loan comes from a policy table.Sat, 31 Oct 2026 00:00:00 GMThttps://ikebude.github.io/blog/reproducible-model-retraining/https://ikebude.github.io/blog/reproducible-model-retraining/Moniguard's metrics used to change on every run. Three small fixes made the retrain reproducible, and CI now proves it on every retrain.Sat, 31 Oct 2026 00:00:00 GMThttps://ikebude.github.io/blog/read-only-sql-guard-for-a-bi-agent/https://ikebude.github.io/blog/read-only-sql-guard-for-a-bi-agent/How my BI agent keeps metrics consistent with a semantic layer, and why its SQL guard parses queries instead of matching text.Thu, 29 Oct 2026 00:00:00 GMThttps://ikebude.github.io/blog/why-my-fraud-model-reports-pr-auc/https://ikebude.github.io/blog/why-my-fraud-model-reports-pr-auc/In Moniguard, about 2.8% of transactions are fraud. Here is why accuracy would flatter a useless model, and what I report instead.Thu, 29 Oct 2026 00:00:00 GMT