Don Karter
CEO & Co-founder, Muvon
I started coding at 12. No computer at home yet — just books. I'd read through them, write code on paper, mentally trace what it would do. When I finally got my cousin's hand-me-down IBM PC, I had months of theory ready to run. That gap between understanding something and being able to execute it — I've been closing it ever since.
From there: the early mobile web (WML, xHTML, before anyone had an app store or knew what one was), then PHP and backends, then years of high-load architecture. Music platforms, fintech, crypto, SaaS, social — the domain kept changing, the problems didn't. Scale, reliability, maintainability. Those transfer everywhere.
Startups along the way. Some worked. Some were very educational. I kept building through all of it.
How I Think
Done is better than perfect. Took me years to internalize this. I'm a perfectionist by nature — still am — but I've learned that good and shipped beats perfect and delayed every time. The 20/80 rule isn't laziness, it's judgment about where effort actually compounds.
KISS and DRY aren't beginner rules. They're the hardest things to maintain at scale. Every system I've seen fall apart did so because someone kept adding complexity instead of stepping back and simplifying. I apply this everywhere — including AI pipelines. You don't need a 12-step RAG configuration to get good results. You need to understand what's actually failing and fix that one thing. Pragmatic over complex, always.
Architecture is a constraint problem. Good systems aren't clever — they're obvious after the fact. The goal is a design where the right thing is the easy thing, where you can hand it to someone new without a week of tribal knowledge transfer. Complexity is always a cost.
Build for the failure case first. Anyone can design a system that works when everything goes right. High-load infrastructure is designed for when it doesn't — traffic spikes, cascading failures, the 3am call you want to avoid. That's where I spend most of my thinking.
AI amplifies decisions, good and bad. I was using AI to write code before most people thought it could write anything useful — before Codex, before Claude. Rough tools, weird APIs, a lot of manual work. But I could see what it was going to become, and I built workflows around it early. That head start shaped everything at Muvon.
Learning never stops. New domain, new stack, new failure mode — I'm genuinely interested in all of it. Healthy lifestyle feeds this: clear head, consistent energy, no burnout spiral. Based in Bangkok now, after years in Phuket. Movement helps.
What I Built and Why
Two tools worth knowing the backstory on.
Octomind came from a real frustration: early AI coding workflows burned tokens at a brutal rate. Context bloat, redundant re-reads, sessions that reset after every exchange. I built something leaner — an agent runtime focused on token efficiency and persistent sessions. It predates most of the mainstream tools. We still run Muvon on it.
Octocode came from drowning in a large codebase. Standard search — even good search — kept surfacing the wrong things. I needed something that understood what I was looking for, not just what words appeared in which file. Built it for myself first, as a human tool. It became an MCP server later, when agents needed the same thing.
Most of what we build at Muvon goes open source. That's a principle, not a marketing decision. I'm also a contributor to ManticoreSearch — a serious open source search engine for workloads that need speed and control.
What Drives This
Muvon started as a reminder as much as a studio. Never give up. Always move forward. It's on the wall, effectively — the name itself is a prompt.
I do IT consulting on backend architecture and AI integration when it's the right fit. Still active, not the main focus.
I care deeply about open source. The best tools I've used were free and built by people who had the same problem I had. I try to contribute back.
Languages & Obsessions
Languages: Rust, Go, TypeScript, PHP (yes, still — not apologizing) Obsessed with: High-load backend architecture, RAG pipeline design, token-efficient AI runtimes, retrieval systems, system design for maintainability Active since: 12 years old. Professionally: 20+ years across every layer of the stack.
Expertise
- Distributed systems and high-load backend architecture
- AI agent runtime design (session persistence, multi-provider routing)
- RAG pipelines — chunking, embedding, retrieval ranking
- System design for maintainability at scale
- Rust and low-level systems programming
- Open source: building it, contributing to it, running a studio on top of it