Background
Client/server systems, full stack.
Enterprise platforms, consumer products, real time applications. Architecture that holds up under real traffic and real teams.
Profile // Systems Architecture
Full stack systems architecture across enterprise and consumer products. Currently researching AI assisted development to separate what works from what's hype.
Background
Enterprise platforms, consumer products, real time applications. Architecture that holds up under real traffic and real teams.
Current Research
Career // Timeline
Red River College, Winnipeg. Two-year diploma in computer analyst and programmer.
Joined a small shop in Winnipeg building inventory management software for restaurants.
Migrated the product from PowerBuilder 11 to .NET Framework 4, both frontend and backend.
Built the enterprise version from scratch, connecting multiple restaurant locations to a head office. ASP.NET MVC on .NET Framework 4, hosted in Azure.
Migrated the product from on-premise packaged software to a SaaS model. Completed successfully and the company was acquired in 2020.
Moved into complex agricultural projects connecting manufacturing systems together, optimizing planning and sales workflows.
First contact with LLMs. Initially saw it as a realized Cleverbot, used it for creative writing. None of the output was good but recognized "this is the worst it will ever be." Learned the difference between context windows and actual learning.
New projects and features built on open source technologies and cloud infrastructure. Picked up GCP, Python, and various JS frameworks alongside existing .NET Framework 4.5 systems.
Started using Copilot autocomplete. Used ChatGPT to learn Python and build early games. Created a Wolfenstein 3D raycaster one-shot benchmark to test new models as they came out. Tested local models with Ollama. Discovered Claude through aider, heavy API spend.
Google conference in Vegas, tested Gemini. Built increasingly complex Python games with aider, then Cursor. Switched to Cursor for cost reasons. Learned React, Next.js became favorite frontend. Most models still failed the raycaster benchmark.
August: shifted from assisted coding into vibe coding. Lost faith in OpenAI, moved to Claude and Google. Reasoning models started one-shotting the benchmark. Heavy Three.js, complex game architecture. Gen AI golden era across all modalities. Spend became unsustainable. Tried Replit, Bolt.new, Windsurf. Started thinking in compute per dollar.
Migrated an internal application from .NET Framework to .NET 10 with a new Vue 3 frontend. A long-term goal I was able to see through. The application can now run in the cloud and be developed on from any platform, with future SaaS potential.
Found Deepseek overhyped but it exposed cracks in the AI hype narrative. Cut services and spend. Built more complex web apps and Three.js game systems. Used premade assets and Mixamo for animations.
NFT projects as hobby. Complex 3D games with usable AI-generated models. API spend surprises on experimental projects. CRUD apps fully commoditized. Claude Code changed everything by freeing workflow from VS Code. Developed a sense for model intelligence: Claude smart, GPT equally smart, Grok creative but dumb, Gemini overthinks.
Integrated systems software for livestock production. Optimizing plant production planning, maximizing sales, full inventory utilization, and logistics tracking.
Four agent tabs running in parallel across projects. ~$150/month on AI services. Using Codex 5.4, Claude Opus 4.6, Gemini Flash 3, Grok, Z.AI GLM 5.1. Not convinced by full agent harness companies. Open source and Apple may have the future advantage.