Lluminor
Available for new work

Developer.

I build things — interfaces, systems, and the layer between them. Design, code, deployment, end to end. Whether I'm the right person for the job is yours to decide. The work is right here.

5x

Hired

Repeat and referral client work

0

WIP

Currently under active build

16

Live

Running and accessible online

26

Projects

Built, tools, and client work

Deep Dives

System Case Studies

Lead Pipeline with Contact History
Lead
Touch
D1
+3d query

The Problem

287 prospects (183 not yet contacted, 55 mid-outreach, 49 blocked) lived in a Postgres table with one column: last_contacted. That answers "when," not "how many times" or "through what." The actual rule I work by — follow up if no reply after 3 days — had to live in my head, checked by opening the table and eyeballing dates.

Architecture

Moved everything into the backend's D1 database. business_leads keeps the same fields the Postgres table had; a new lead_contacts table adds one row per touch — channel, template, outcome, timestamp — instead of overwriting a single field every time. lluminor-admin got a leads section on top: list, filter by status, per-lead contact history, log-a-contact action.

Trade-offs

This is a join instead of a column read. For 287 rows it's free; I'm not pretending it wouldn't need an index at 50k. Took it anyway because the column version physically cannot answer "how many times has this person been contacted" — that number doesn't exist without the extra table.

Business Impact

MAX(contacted_at) + 3 days per lead is now a query, not a memory. contact_count comes from counting real rows, so it can't quietly drift the way a manually-incremented field does. And I stopped opening psql to check who to email next.

Real Bug, Real Fix

Something actually broke.

Not a hypothetical — a real production incident, and the real fix for it.

The 429 that looked like a network outage

The admin dashboard would randomly fail to load data. Turned out to be a request-ordering bug in the server's middleware — fixed in one line, with a test added so it can't silently come back.

Interactive · Live

Recent Activity

Pulled live from GitHub's public events API, right now — not a snapshot. This is whatever I actually pushed most recently, whenever you're reading this.

No recent public activity to show.

Technical Stack

What I Actually Build

A quick scan of the toolset — the deeper claims live in the case studies above, backed by real systems, not repeated here.

01

Frontend Engineering
ReactTypeScriptState managementComponent systems

02

Animation & Motion
GSAPScrollTriggerThree.jsPage transitions

03

Design Systems
Tailwind v4CVAshadcn/RadixMonorepo UI packages

04

Infra & Deploy
Cloudflare WorkersSPA routingEdge cachingCI/CD

Workflow

How I turn ideas into deployed reality.

01

Inspiration

Pure curiosity — no problem to solve, just something I want to exist.

02

Research

Understand what exists, what failed, what's been tried. Skipping this compounds later.

03

Plan

Define the smallest slice that proves the idea. One feature, one route. No architecture phase.

04

Prototype

Fast and disposable. The goal is to learn what works, not to write code that lasts.

05

Harden

Types, error handling, edge cases, deploy pipeline. This isn't the prototype — it has to survive real use.

06

Ship

Deploy it. A real URL, not localhost. If it's not accessible, it doesn't exist.

Let's collaborate

Ready to build something incredible?

Open to collaborations, interesting projects, and anything that requires both design and engineering. Let's make the web a little more beautiful.

Jamil Harun · 2026