Dilusha Gonagala Software engineer Writer
I’ve been making computers do things since 2004, when my first one ran Windows 95 on around 64 MB of RAM. Professionally since 2012. This site is the long exposure: what I’ve built, what I’ve written, and 2,121 thoughts I keep coming back to.
How to read the sky: one ring per year
The observer
That feeling of wanting something to work never really left. This site is where I think out loud about the work, along with books, kindness, family, curiosity, and the life happening around it.
A little more about meWindows 95, around 64 MB of RAM, and 22 years of wanting to know how things work.
Around 17 hours and roughly 2,000 linked slides. I didn’t know what programming was yet. I only knew I wanted it to work.
How I fell in love with code14 years since, across manufacturing, e-commerce, financial systems, and native apps.
A desk at my first job as an Associate Software Engineer. No AI, no shortcuts.
A reminder from 2013Still building, still writing, and still curious enough to lose track of time making something work.
Star chart
Software I’ve built, unlikely beginnings, and questions I haven’t finished asking. Pick a point and wander.
Built by me / AgentsEye
A native Mac app for usage limits, local history, reset windows, and alerts across the major AI coding providers.
The difficult part is making different providers tell one coherent story. A fresh quota window, a stale reading, and continued usage cannot be treated as the same event.
Built by me / Links on Link
A home for digital products, branded links, custom domains, and campaign analytics.
I built the system from the public page through payment confirmation and delivery. It does not have real users yet, so I treat it as built work, not a validated business.
Built by me / MemoryCrib
A thinking studio for curious kids, built around writing, logic, puzzles, and memory palaces.
I’m building the experience and API with Angular and NestJS. Spatial memory is one way into a bigger question: how can a child practice an idea and find their way back to it? Currently in development, with a public waitlist.
Built by me / DeSlop
A writing tool for removing filler without handing the final decision to a model.
Local cleanup accepts drafts up to 20,000 characters. Model-backed cleaning is limited to 12,000 characters, keeping those requests deliberately bounded.
From the notebook / Where it began
My first computer, in 2004. That’s where my engineering journey began.
By around 2006, I was making a calculator out of roughly 2,000 PowerPoint slides, before I understood what programming was. Seventeen hours later, it worked.
From the notebook / The universe
What can we know about the universe, and where does our imagination take over?
My notebook has room for speculative questions as well as software. This essay follows a late-night thought about black holes; it’s personal curiosity, not a scientific claim.
From the notebook / Mental models
What would have to be true for your plan to work? What happens if it isn’t?
A different question can expose an assumption you stopped noticing. The note library has prompts for second-order effects, incentives, reversibility, and the room a system needs when things go wrong.
From the notebook / Being human
Including the way you talk to yourself while you’re figuring things out.
The notebook is also where I write about patience, kindness, and the life happening around the work. You don’t have to be here for software.
Engineering range
I move between the system and the screen. The tools depend on the work.
Platforms & infrastructure
AWS infrastructure, repeatable environments, deployment pipelines, and production troubleshooting. I use Terraform and CI/CD for web systems and native release paths.
Backend systems
Node.js, TypeScript, and NestJS. APIs, payments, background jobs, data access, and performance work. This includes financial processing where deadlines and system limits are part of the design.
Web & native applications
React, Angular, and Astro on the web. Swift and SwiftUI on macOS. React Native and Expo on iPhone.
AI experiments
Model integrations, output checks, and evaluation of generated text. I also experiment with Python, LoRA, and QLoRA. These are experiments, not claims that the hard parts have been solved.
Experience
Five systems. Five different things that could not be hand-waved.
Financial systems / Reliability
Statement processing began a few hours after the billing cycle opened. That left roughly three to four hours for each batch to finish.
Slow queries and hard system limits were eating into that window. I worked with the team to find the bottlenecks, improve the existing processing path, and rewrite the parts where smaller changes were no longer enough.
The decision was not simply whether to optimize or rewrite. It was deciding which parts justified each approach while still working inside limits we could not remove.
Company details, volumes, and implementation specifics stay private.
AgentsEye / Integration & reliability
AgentsEye works across the major AI coding providers. Each one represents usage limits, resets, and unavailable data differently.
I built the logic that distinguishes fresh, stale, and unavailable readings, tracks usage across reset cycles, and avoids repeating the same alert.
History stays local. Native widgets and signed updates make the app useful after the dashboard is closed.
See itLinks on Link / Commerce
A storefront, a Stripe payment, and a downloaded file all need to agree about what happened.
I built across the creator dashboard and public storefront, including custom domains, branded links, verified payment notifications, digital delivery, and campaign analytics.
The product does not have real users yet. I include it here as an example of the system I built, not as evidence of traction.
See itDeSlop / AI product quality
DeSlop accepts drafts up to 20,000 characters. Model-backed cleaning stops at 12,000.
I worked on the checks around those rewrites: comparing the result with the source, protecting concrete details, and leaving the result editable.
Fluent output is not proof that the meaning survived.
See itFastRight / Mobile & data continuity
FastRight is in private beta. Its timers need to survive app restarts, data-format changes, and releases without quietly changing someone’s history.
I used saved timestamps, versioned migrations, and shared app and widget state. Before a backup replaces existing progress, the app validates it and keeps a recovery copy.
It works without requiring an account.
Private beta.
The notebook
Some are about software. Some are about life. Some made more sense when I wrote them. The note in the sky holds 2,121 more.
Open the notebookIn 2013 it was just me, Sublime Text, and curiosity. No AI, no shortcuts. Just grind. Looking back now, I see how every small step shaped the life I have today.
Scarcity teaches grit in ways abundance never will. Looking back, my best skills were forged when every click and every second online had a cost.
George Bernard Shaw once wrote that there are two great tragedies in life: the first is not getting what you want, the second is getting it. He was right.
I also take on a small number of focused, part-time freelance projects. If you have something specific in mind, contact me.
