My Stepbrother's Dirty Little Secret novel Chapter 101

101 /103

Raphael

━━ ⛓ ━━

The Midland Trust team was early and that told me everything I needed to know. Banks only showed up early when they were bleeding.

I stepped in first. I gave no greetings beyond a short nod. My team followed behind me like a shadow, disciplined, silent, in sync.

I took my seat at the head of the table, set my laptop down, and connected to the display. The Orion Vector interface lit up the main screen.

Across from me sat Midland’s people. CFO. Head of Risk. Chief Compliance Officer. Two analysts who looked like they hadn’t slept in days.

“Let’s not waste time,” I said, voice even. “You’ve already lost thirty million. The only question is how much more you’re planning to lose before you fix it.”

The CFO leaned forward slightly, “We’re here to understand how your system prevents that.”

I glanced to my right, “Daniel.”

Daniel was my Lead Machine Learning Engineer, late thirties, sharp mind and built the base architecture of the model we were selling today.

He stood, clicking his remote, the screen shifted to graphs, movement patterns and risk curves.

“This is the current fraud detection timeline,” he began, “Your system flags transactions after execution. That’s reactive. By the time you see it, the money’s gone,” he clicked again, “Our model shifts that window forward.”

The chart changed. Earlier spikes. Predictive markers.

“We don’t wait for fraud to happen,” Daniel continued, “We identify behavioral deviation before the transaction clears.”

He spoke well but this part wasn’t the sale. It was the setup. I leaned back slightly, eyes moving, not to Daniel, but to Gianna. She was standing near the screen now, hands lightly clasped in front of her. Daniel wrapped up the architecture overview and sat down.

The Head of Risk spoke next. “Prediction models always come with false positives. How are you controlling that?”

I tilted my head slightly towards Gianna. It was her cue.

She stepped forward, “The system doesn’t use fixed thresholds. It adapts based on user behavior clusters. So instead of flagging every anomaly, it scores intent based on pattern deviation.”

She didn’t rush that part.

She pointed at the screen, fingers steady.

“Like this,” she continued, “If someone logs in from a new location, that alone isn’t enough. But if it happens with a new device, at an unusual time, and with spending behavior that doesn’t match their profile—”