I'm an indie maker with a background in large-scale systems, now building small, focused Chrome extensions and practical, offline-first tools.
I care about clarity, reliability, and tools that quietly fit into real workflows.
I design tools the same way I design systems: clear boundaries, predictable behavior, and no hidden complexity.
Most of my work lives inside the browser and focuses on:
Turn any PDF into clean, readable Markdown – for people and for AI agents.
Handles messy, real-world documents, not just perfect ones. Use it as a web app, a browser extension, a REST API or a hosted MCP endpoint.
Notes that live directly on any website.
Floating Notes is a Chrome extension that lets you capture thoughts, links, and context exactly where they happen – without switching tabs or breaking focus.
Notes can be tied to:
Designed to stay out of the way until you need it.
Multiple parallel timers and alarms in Chrome. Named timers, presets, group launch – offline, no accounts.
AI-powered journaling with metaphorical cards. Draw a card, reflect with guided templates, get a clear next step in 2-7 minutes.
Send files directly between devices, end-to-end encrypted. P2P via WebRTC – no cloud, no sign-up, no size limits.
Two-way converter: turn any image into a Base64 / data URI string, or decode Base64 back into an image. SVG supported. No uploads, no network – everything runs locally.
After years of building large systems, I've learned that most tools fail not because they lack features, but because they interrupt the user.
I believe good tools:
These dashboards and the architecture map aren't slideware. They're generated straight
from each repository by tooling I built – a metrics pass over the git tree
plus a live snapshot from production. The numbers here are the numbers in the source.
System architecture – many entrypoints, one entitlement-aware job core. Click to zoom.
Article on Substack
"Add a PDF upload form" sounds like a 30-minute task. But a working upload and a secure upload are two different things. I broke down 7 real attack vectors on file upload and showed how we mitigated each one in a Go backend – with real code and honest trade-offs.
Article on Substack
What dog training unexpectedly teaches a manager who works with AI agents: about signal quality, safe autonomy, and the cost of literalness. Five lessons that changed the way I work with agents and assign work to people.
I occasionally write about building tools, architectural decisions, and lessons learned from shipping and maintaining real products.
Selected articles:
dimlight is my personal space on the internet.
It's where I share what I build, experiment with ideas, and publish tools that don't fit into big platforms.
You may also see me online as dim0802.