Give me a hard problem, an unfamiliar stack, or a topic I know nothing about yet, and I'll happily disappear into it until I understand it, then build something with it. That curiosity is what DevButler runs on: precision engineering delivered directly by the person who wrote the code.
I've been writing software for years, and the part that's never worn off is the curiosity, a new language, an unfamiliar protocol, a bug that doesn't make sense yet. Give me something I don't understand and I'll happily lose an evening pulling it apart, then rebuild it better.
That's the energy I bring to client work too. Whether it's a custom application, a website, or wiring up AI to automate something tedious, I care about getting the details right clean architecture, code that lasts, and software that actually solves the problem instead of papering over it. I use AI heavily to move faster on the routine stuff, but the decisions that matter the ones that determine whether software holds up still come from experience, not a model. No sales team, no project manager in between: you work directly with me.
The same attention I bring to my own side projects, I bring to yours.
New languages, frameworks, and topics if I don't know it yet, that's usually reason enough to dig in, instead of reaching for whatever's already familiar.
Every line of code is reviewed, tested, and built to last. I take pride in getting it right the first time no shortcuts, no spaghetti code left behind.
You work with the person actually writing the code clear communication, realistic timelines, and zero surprises on your end.
A sample of what I've built some for clients, some just because I wanted to learn something new.
A 3D browser MMO brawler built in BabylonJS. Punch and kick your way through Influencer enemies across a gym and outdoor world, level up by landing hits, and get visibly bulkier as you fight. A Go backend speaking WebSockets is in active development to broadcast player positions, animations and combat events in real time across a shared world.
A marketing platform that takes the grind out of running LinkedIn and Facebook ad campaigns. A Claude Haiku 4.5 agent analyses creatives, recommends audiences, and writes headlines inline, so the marketer stays in flow instead of pinging between docs and external APIs.
Case Study
A desktop app for spinning up local AI agents that actually do work on your machine, running shell commands, editing files, building projects end-to-end, entirely against locally-running Ollama models. Nothing leaves your computer. This is only V1, we are planning more features to improve LLM capabilities locally.
Notes on what I'm building, what I'm learning, and how I think about software.
I built a local AI agent playground for Ollama and ran the same prompts through Qwen, Phi4, Mistral, Llama, Gemma and DeepSeek. "Tool calling" turned out to be six different things wearing the same name, and the lessons learned along the way were the most fun part.