Paraiso Packing System

PARAISO PACKING · 2023 · OPERATIONS & CLIENT PORTAL

Paraiso Packing System

Minutes, or nothing at all. Now one third of a second.

The system a Los Angeles labeling company runs its warehouse on, and the portal its clients log into to find their merchandise. Staff had stopped opening the reports that never finished.

// 01 · THE SITUATION

The software had become the bottleneck.

Paraiso Packing labels and ships merchandise for apparel brands out of the Los Angeles Fashion District. One system runs all of it: what arrived, what's been labeled, what's going out, who gets invoiced. Their clients log into the same system to see where their merchandise is.

By the time we were brought in, the reports at the center of it took minutes to open. The vendor invoice page didn't open at all: twelve seconds, then an error. Staff had stopped opening the worst of them and worked around them, and bringing on a new admin had become nearly impossible, because the tool you'd train them on crashed half the time.

The cause was not subtle. The system loaded every record it had, every time anyone opened a page. That works at a few hundred rows. Paraiso had tens of thousands, which is not a large database. It's an ordinary week for a business this size. We didn't build it. We were brought in to make it work.

Meanwhile a client logging in to see where their merchandise was waited, gave up, and picked up the phone. Which is the call the portal exists to prevent.

// 02 · THE FIRST THING WE BUILT

A copy of the original, untouched, on its own server.

Before we changed a line, we stood the application up a second time, exactly as we found it, and left it running.

On an inherited system it's genuinely hard to know where the existing problems end and the new work begins. A running copy of day one gave everyone the same reference point. When something odd surfaced, we opened both and looked. If it was already there, it went on the list to be prioritized like everything else, instead of quietly absorbing time that had been budgeted for something else. It kept the scope of work honest in both directions, which serves the client at least as much as it serves us.

It also became the best demonstration we had, the vendor page today beside the vendor page as it was. And it's the reason every "before" figure on this page still exists, measurable by anyone who asks.

// 03 · WHAT WE BUILT

Five fixes the warehouse felt.

// 01 · THE REPORTS

9.91 seconds to display a single page. Or not at all.

The reports are the system: daily receipts, delayed shipments, shipping, vendor invoices. Everything anyone needs to know lives on one of them, and every one pulled the entire database to show twenty rows. On a good page, the server spent 9.91 seconds building it. On the vendor invoice page, you got the error.

They now load a page at a time, sort on any column, and open in three tenths of a second.

You also stop losing your place. Search an invoice, open it, close it. Before, that landed you back at an empty search box over an unsorted list, and people found the same record three and four times in a morning. Search, filter and page position now survive every action, and they live in the web address, so the back button works and a link to page four opens on page four.

// 02 · SHIPPING SUBMISSIONS

It didn't just fail. It left you guessing what had saved.

Every day an admin records everything coming in and going out. That submission ties a client's inventory to a tracking number, and it's what a client sees when they ask where their order is.

It failed regularly, and it failed partway. Hundreds of separate saves, one after another, and when it stopped in the middle nobody could say what had gone in. Duplicates? UPS labels paid for on shipments that weren't recorded? Start again and make it worse?

It's one atomic write now. The whole day records together or not at all, and the same query work that sped up the reports made the submission itself far quicker. Nobody stands at a screen at four in the afternoon wondering what they've already paid for.

// 03 · THE NAVIGATION

Nobody should need a manual to use your main navigation.

The whole thing was three dropdowns: Add, Receipt, Report. Twelve screens filed by verb rather than by subject. Label Inventory lived under Add, Delayed Report under Report, so finding anything meant first deciding what kind of verb it was, and the only way to learn what a screen actually was involved opening it. Opening it took minutes.

We were deep in this codebase for months and still had to stop and think about where things were.

We went through it with the client and cut it to five, named after what the business has: Invoices, Shipping List, Vendors, Clients, Label Inventory. No dropdowns. Everything one click from anywhere. It's a large part of why their words for training someone new went from "nearly impossible" to "a breeze."

// 04 · UNTANGLING THE TEMPLATES

Every change used to cost twice.

Every screen existed twice: a desktop version and a separate phone version, drifted over the years into different features, different layouts and different bugs, with the styles and scripts buried inline in each.

Long before we arrived, that meant any change to this system was two jobs. Find both copies, edit both, and hope the desktop edit didn't unsettle the phone version. It's how small requests turn into two-week requests, and it's part of why the system had stayed in the shape it was in.

We rebuilt it as one screen per page that works at any size, with the styles and scripts in their own files. A change is now made once and lands everywhere. Every request Paraiso makes costs less than it used to, and it will keep costing less for whoever works on this system long after us.

// 05 · LABEL PRINTING

A hundred hours a year, back.

Printing the day's shipping labels meant opening each shipment record, clicking in, and choosing print. One at a time, dozens a day, twenty to thirty minutes of somebody's morning, every working day. Across a year that's roughly a hundred hours spent printing labels one at a time.

Now you filter to the shipments you want, select them, and print the lot. That fires dozens of calls to UPS behind the scenes, so it takes about forty-five seconds, behind a progress screen that says what it's doing instead of a frozen page you're afraid to click twice.

Forty-five seconds, once, in place of the twenty to thirty minutes.

// 04 · THE HARD PART

The answer wasn't more compute.

When software gets slow there's a fast answer available: throw horsepower at it. The developers before us had already reached for it. Paraiso was sitting on a large AWS instance, and the pages still took minutes or failed outright. They were paying more every month and waiting exactly as long.

That's what more compute buys when the cause is somewhere else. So we went looking for the cause. It took three tries.

First, the reports stopped pulling everything. The obvious fix, done in the first month: load what's on the screen, not what's in the database. It helped, and the problem was still there, so we dug deeper.

Then we found that displaying one row cost three trips to the database. To show a single line of a shipping report, the system went back three or four separate times: the invoices attached to it, the box count, the tracking number. Fine for one row. Three or four hundred requests for a hundred-row page. And nothing in the code looked wrong, because each request reads like an ordinary line, which is why it had survived. We rebuilt the reports to ask for everything up front and work from what they already have. Three or four requests per row became one request for the whole page.

Then we shipped it, and a week later it was slow again. Our testing was clean. Pages that had taken minutes were coming back in a second or less. Then the client said the slowness had crept back. We were surprised, but they were right, and we could see it on production too.

We went into the server logs and read until we found it: a scheduled job, running every sixty minutes, reading the entire invoice table to flag late shipments. Reasonable when it was written against a few hundred invoices. Against tens of thousands, it meant the server spent part of every hour on a question nobody was asking, and everyone's work slowed while it ran.

We took the time to understand what it did and why before touching it, because a scheduled job you don't understand is one you delete and then find out on a Friday what it was for. Then we rebuilt it to check only what changed.

Two years on, with the business bigger every day since, every page still comes back in a third of a second or less.

// 05 · THE NUMBERS

What it actually did.

60–90 min

returned every day, across three admins

10%

fewer calls from clients asking where their merchandise is

9.6 sec

cut from loading a report

12.4 sec

cut from the vendor invoice page, which now loads instead of erroring

24

full reads of the invoice table every day, eliminated

20–30 min

of the daily saving from label printing alone

“It already does seem quicker since Friday. Less delays were noticed immediately especially in the shipping portion of the site.”

— Grecia Alarcon, Paraiso Packing

// 06 · THE STACK

Why these choices.

LaravelPHPBladeEloquentUPS API

We fixed it in Laravel, because that's what it was written in. The first move a lot of vendors make with an inherited system is a rewrite pitch into whatever stack they prefer, which trades a working system and the client's budget for the vendor's comfort. This business had already invested in this codebase, PHP is squarely inside what we work in, and the fastest route to a fast system was the one already running. It's still Laravel today, which means anyone Paraiso hires next already knows their way around it.

Reports ask for everything up front. Rather than fetching related records while the page displays each row, a report collects what it needs in one request and works from that. This one change removed most of the load time.

Your place lives in the web address, not on the server. Search, filters and page position are written into the URL. Back and reload behave the way a person expects, a link carries its state with it, and two people on two machines never fight over one shared memory of where somebody was.

// NEXT CASE STUDY

National Grid

// START HERE

TELL US WHAT
YOU'RE BUILDING.

Twenty minutes with a senior engineer, and you'll leave with how we'd build it, roughly what it takes, and what we'd worry about first. We read every message ourselves and reply within one business day.

A rough idea is enough to start.

0/1000

Attach a brief, deck, or screenshots (optional)

Allowed: PNG · JPG · PDF · DOC · DOCX · PPT · PPTX · KEY

Up to 5 files, 25 MB total.

Or just email us at hello@queencity.io

// THE CALL

Twenty minutes. A real answer.

You'll leave with how we'd build it, what it takes, and what we'd worry about first.

Loading availability…