About
Paradigm Coding Solutions is a one-person engineering practice with an R&D lab attached. The practice takes on AI and ML work for clients. The lab, Project Nexus Code, builds the tools and the techniques the practice uses next. The two halves are meant to feed each other: client work pays for the research, and the research keeps the client work from getting stale.
Who runs it
Drew Mattick, Primary Developer, Paradigm Coding Solutions.
Drew is the engineer on every engagement. "Primary Developer" is a deliberate title: not "founder," not "principal," not "CEO." It says what the role actually is on the project, which is the person who reads the code and ships the code. A one-person studio is upfront about being one person, because pretending otherwise is how trust gets lost.
Background: AI and ML systems engineering, with a research thread on safety, alignment, and the ethical development of electronic intelligences. Drew is working toward a PhD in computer science, focused on AI/ML research. Engineering before the studio centered on decision-support and reporting UX for leaders in healthcare and manufacturing, the kind of work that compresses the gap between "the data is in the warehouse" and "a leader made a call." That data discipline carries forward into how we build ML systems now.
US provisional patent application filed April 2026, on an autonomous mobile perception platform for building management.
What we use
In production
Running in a deployed system today.
- Applied
- Anomaly detection, explainable risk scoring, and decision-support analytics (the live demos).
- Web and APIs
- TypeScript, Node.js, Astro, Next.js, React, Tailwind CSS.
- Data
- PostgreSQL, Neon, Drizzle ORM, SQLite.
- Auth and email
- Signed-JWT single sign-on across apps, Resend.
- Edge
- Cloudflare Workers, Cloudflare Access, Cloudflare Tunnel.
- Infrastructure
- Docker, a self-hosted server behind Cloudflare Tunnel.
- Quality and ops
- Playwright end-to-end tests, Vitest, GitHub Actions CI.
Also hands-on with
Real experience, not currently carrying production traffic here.
- Models and agents
- Anthropic Claude (the daily build tooling), Gemini and OpenAI-compatible APIs, Model Context Protocol servers (the Lumen demo ships one), multi-agent orchestration, RAG.
- ML and research
- Python for tooling and research code, and PyTorch and Hugging Face for the lab's training runs rather than a deployed service.
- Platform and edge
- Kubernetes, Azure, Hyperdrive, Cloudflare D1, Terraform.
- Web and auth
- Hono, SvelteKit, Better Auth.
- Payments and ops
- Stripe (test mode in the demos), Sentry, PostHog, Turborepo.
- Languages
- C# and .NET when it earns its place. Rust and FastAPI where they fit.
The split is the point. One flat list lets a reader assume everything on it is running somewhere, and that is the kind of claim nobody ever has to defend. "Use everything" is not a stack, it's a confession.
The research thread
A real part of what we do is the ethical development of electronic intelligences: AI safety, alignment, and the questions about how autonomous systems should and shouldn't be built. This isn't a manifesto and it isn't a marketing posture. It's a working commitment that shapes the engineering: contract-first agent workflows where a person approves every merge, a written record of each change and why it was made, and deferral to humans on the decisions that should never be automated. Provenance, on the work page, is where it goes next: a tamper-evident ledger of the decisions agents make.
We talk about it on the about page instead of burying it because a client deserves to know how the studio thinks about the work it's being hired to do. If your project would be safer in slower hands, we'll say so.
Where to find us
Based in the greater Milwaukee area, Wisconsin. Reachable through the channels in the footer and through the contact page.