I build systems that run themselves.

Chart signals become bracketed orders across a fleet of simulation and evaluation accounts, inside a cage of deterministic guards, with AI agents above the signals managing exposure and risk posture on earned authority. Every incident gets a fault line — AI, system, or me — on the journey page. Simulation and prop-firm evaluation accounts only; not advice.

A live futures dashboard where support and resistance are ranked by how often each level actually held that day, strategies are scored net of real costs, and a confluence engine only draws a signal when independent evidence stacks — then explains why. Zero AI at runtime: models helped design it; deterministic code ships it. On the ventures page.

A device ships straight from Dell, HP, Lenovo, or Apple to an end user, powers on, and configures itself — apps, policy, security baseline, identity — with no IT hands on the hardware. I've built these flows with vendor integrations on both the Windows and Apple sides, replacing entire imaging benches.

Reusable PowerShell that deploys a complete CIS-aligned Intune/Entra configuration to a fresh tenant — proof-of-concept in hours instead of weeks, and the same code carries to production. The portal is for reading; the pipeline is for writing.

Moving organizations from on-prem SCCM to cloud-native management: co-management, CMG for internet-based clients, hybrid-to-cloud-only conversions — including the VPN, firewall, and Conditional Access redesign nobody budgets for.

Event-driven automation with self-healing properties, and LLMs pointed at the ugly parts of operations — log triage, packaging drafts, policy generation, incident autopsies — always with a human owning the blast radius. The thesis lives on the ventures page; the lessons land in the field notes.

The toil was provisioning and configuration: imaging benches, repeated policy builds, hand-written detection rules. The intent got encoded as baseline-as-code — PowerShell libraries that stand up a full CIS-aligned Intune/Entra configuration from nothing. AI accelerates the drafting and the log triage; Autopilot, compliance policy, and staged update rings do the enforcing.

Result in shape, not in numbers: a device ships from the vendor to a user, powers on, and configures itself correctly — with no IT hands on the hardware, and a documented way to roll any of it back.

The same method, aimed at a personal engineering problem: making a noisy, judgment-heavy decision process legible. NQ Terminal is the result — a live futures dashboard where support and resistance are ranked by how often each level actually held that day, strategies are scored net of real costs, and a confluence engine only draws a signal when independent evidence stacks, then explains why.

The tell that the method held: zero AI at runtime. Models helped design and harden it; the shipped tool is deterministic Python and a single HTML file, and when nothing qualifies it honestly shows nothing. Built as an analysis tool, not advice — see the disclaimer below, and the ventures page for the live link.

Both systems above have the same friction: a human still has to translate intent into the tool's dialect. The venture takes the method one layer up — you state what you want in plain English, the model drafts it as validated policy-as-code, it stages to a pilot ring, and every change arrives with a readable rationale and a one-step rollback.

Same division of labor, made general. More on the ventures page.