I Build Systems That Solve Problems Others Work Around

The problem

Outpatient surgery centers had a consistent, unaddressed need. Physicians don't want to deal with broken hardware, network outages, or software lockouts. They want IT to work, and when it doesn't, they want someone physically there to fix it. When I joined Integrity HIT as Director of Business Development, the company was an IT staffing agency with no on-site capability. The opportunity was clear.

What I built

A full IT field services practice built around putting skilled technicians on site so physicians never have to think about IT. Network closet construction, Cisco rack-and-stack, structured cabling, fiber runs, server configuration. The scope covered the full facility lifecycle, from de novo buildouts through ongoing managed services.

The outcome

It grew into Integrity's largest revenue-generating service line, with over 100 onsite projects executed in the first year.

The problem

The field services practice was growing, and that growth created a new problem. Too many clients onboarding at the same time, too many jobs competing for the same technicians, and no clean way to see who was available when. The service itself was solid. The coordination behind it wasn't keeping pace, and that was starting to affect quality.

What I built

An AI agent using n8n for workflow orchestration, Jobber for field service management, and the OpenAI API for intelligent routing. It handles client intake, job scheduling, and invoicing across the field services operation. It matches technicians to jobs based on availability, triggers invoices on completion, and keeps the pipeline organized without someone manually tracking every moving piece.

The outcome

Operational efficiency went up by over 70%, and the job pipeline grew by over 30%. The bigger win was that the team stopped spending time on logistics and went back to spending it on the work itself.

The problem

MediReport is a European cardiovascular information system that I helped bring into the American healthcare market. As they started implementing across outpatient OBLs, ASCs, and clinics doing cardiovascular procedures, the project management side couldn't keep up. Tracking was spread across Excel spreadsheets, multiple OneDrive instances, email chains, and weekly PowerPoint status decks. Different people were updating different files. Version conflicts were constant. There was no single source of truth, and bottlenecks were showing up at every stage.

What I built

A React + Firebase web application that replaced all of those disconnected systems with one tool. It includes customizable implementation plans per site, role-based access for internal teams and customers, and an AI assistant that lets project managers log updates in plain English. Someone can say "I had a call with this vendor, there's going to be a delay interfacing these modalities," and the AI routes the information to the right fields, flags whether the task needs clinical or engineering involvement, tracks deadlines, and sends reminders.

The outcome

Implementation efficiency improved by over 20%. That might sound modest, but these implementations typically take 8 to 16 weeks. Shaving 20% off a three- or four-month rollout means weeks of time recovered per site. The MediReport team now runs all of its US implementations through this single platform, and it's an active Constance IT engagement.

The problem

AI-generated content was taking off, and most of the conversation around it felt like hype. My girlfriend runs a wellness brand with regular sponsorship offers, so I had a real use case ready. But before building anything for her, I wanted to understand every step of the pipeline myself.

What I built

The full pipeline from scratch, starting with automated YouTube content as my test case. You give it a one-line premise, and the system generates a scene-by-scene script, produces voiceover narration per scene, generates AI-animated video clips, and assembles everything into a finished video with word-synced captions, ready to publish. I built it on n8n for workflow orchestration, Claude API for script generation, ElevenLabs for voice synthesis, Veo 3.1 and Kling 3.0 for video generation, Creatomate for final assembly, and Blotato for multi-platform publishing.

The outcome

Once I understood every step, I adapted the pipeline for my girlfriend's brand. She's now producing and publishing AI-generated UGC at a pace that wasn't realistic before. For me, the value was less about the pipeline itself and more about developing a clear picture of what AI content creation can actually do and where it falls short.

The problem

Prediction markets were taking off and people were claiming you could set up a trading bot overnight. I wanted to find out for myself what it actually takes to build one of these and whether there's a real edge behind the noise.

What I built

I started from first principles. I researched how weather data is measured, where weather stations source their information, and how the data gets produced and hosted. I selected an AWS data center co-located in the same cluster as NOAA's weather data to minimize latency. Then I built the trading system in Python, using NumPy for the statistical modeling: it ingests real-time NWP ensemble forecast data from NOAA, constructs empirical probability distributions with station-level bias correction, and executes trades on prediction exchanges with built-in risk controls like exposure ramps, drawdown circuit breakers, and per-market caps. It runs autonomously on AWS EC2, processing over 200 markets per cycle across 20 US cities, and sends me automated performance reports.

The outcome

The system is fully operational. More useful to me, though, is the understanding I built along the way. The data engineering, the statistical modeling, the infrastructure decisions. The gap between the hype and the reality is significant, and the only way I could have known that was by building the thing.

The problem

The two people in my life who cook the most had recipes saved in a dozen different places. Handwritten notes, Instagram saves, screenshots, bookmarked websites, physical cookbooks. They could never find what they needed when they needed it.

What I built

Bella, an AI-powered recipe manager for iOS, built in Swift and SwiftUI. You can import a recipe from anywhere: take a photo of a handwritten card, paste a web link, upload a screenshot. The app uses the OpenAI API to interpret handwritten notes via image recognition, and to extract structured recipe data from web pages while stripping out ads, navigation, and filler content. Everything gets formatted cleanly and saved to your account. I'd never done iOS development before this project. I took it from concept to App Store launch on my own.

The outcome

It's live on the App Store and used by my girlfriend, my mom, their family, and friends. It's the simplest project here, but it might be the most telling. I saw a problem, learned a new platform, and shipped a real product.