Systems built for the work behind the work.

The problem

A growing agency creates operational complexity quickly. Sales, campaigns, creator relationships, analytics, reporting, proposals, and client delivery all produce data, decisions, and handoffs. When those live in separate tools, the team becomes the integration layer.

What I built

• A connected CRM and operating data model across brands, creators, campaigns, deals, deliverables, and reporting.
• Internal and client-facing workflows for creator analytics, profiles, proposals, media kits, portals, and recurring reports.
• Source tracking, approval gates, and production safeguards for AI-assisted work.
• Cloud delivery and lightweight tooling shaped around the team's actual operating habits.

The result

A calmer operational layer with clearer ownership, more reliable source data, and fewer manual checks hiding between tools.

The problem

Creator data arrives at different times, in different formats, and at different levels of detail. Copying numbers into a polished report is easy. Knowing whether the number is current, comparable, and supported by evidence is the harder problem.

What I built

• Structured intake for LinkedIn, YouTube, newsletter, and campaign performance data.
• Clear source and freshness rules so reported numbers can be traced back to evidence.
• Reusable profile and reporting outputs that stay polished without hiding uncertainty.
• Automation for the repetitive parts, with human review where judgment still matters.

The result

Faster reporting with a more reliable path from raw evidence to the version a client sees. The team spends less time moving numbers and more time understanding them.

The problem

AI can make a workflow faster while quietly making it less trustworthy. The failure is rarely the model alone. It is usually unclear source data, missing review, weak ownership, or an action that happens too early.

My approach

• Start with the business decision and the source of truth, then decide where AI is useful.
• Separate drafting and recommendation from actions that change live systems.
• Keep evidence, confidence, and missing information visible to the reviewer.
• Use deterministic automation for repeatable rules and AI for work that actually needs interpretation.

Where it is used

Internal reporting, proposal preparation, creator research, operational triage, CRM cleanup, and client-facing production work. The system is designed to help a capable person move faster without asking them to trust a black box.

The problem

Middle click is useful for opening links, closing tabs, and navigating certain apps, but a MacBook trackpad does not provide the gesture naturally.

What I shipped

• A native menu bar app with three-finger and four-finger options.
• Conflict detection for macOS gestures that can interfere with the app.
• A lightweight build with no Electron runtime or unnecessary background services.
• A signed, notarized release with public source code and a simple product page.

View it