All work

Daily sorting reports and hours for a containment-sorting crew at IQ Automotive Services.

A skid of eight cartons on a pallet, with its supplier label pulled off and shown to the right inside a camera viewfinder. Callouts mark the part number, supplier and serial barcodes, and note that one photo is read in the browser and the skid is filed under its part. Part no. (P) CX418DMOAA02 Supplier (V) 000417220 Serial (S) 104882317 One photo Read in the browser Skid 7 of 12 Filed under its part
For
My crew at IQ Automotive Services
My role
Built it, and use it every shift to write the reports
When
September to October 2026
Used by
Site supervisors at the end of a shift; IQ’s office for reports and hours
Status
Used every shift since late September 2026
Built with
Next.js 16, Supabase, zxing-wasm, Claude Haiku, ExcelJS, jsPDF, Vitest
CH_01

The floor

Every number lived in more than one place.

When a supplier’s parts are suspect, a containment crew sorts every skid at the customer’s plant before the parts go on the line. We check each part, tag the rejects, and at the end of every shift report what was sorted, what was rejected, and who worked how long.

That reporting was Excel. One workbook per part, filled in by hand from the skid labels and paper notes. Hours typed again into a daily hours sheet, then again into a monthly one.

I work on this floor and I write these reports, so I built the tool I wanted at the end of my own shift.

Places each hour was typed before: the workbook, the daily sheet, the monthly sheet
3
Page for the whole end of a shift, top to bottom
1
Hours the app moves between people on its own
0
In use every shift since late September 2026
Daily
CH_02

One photo per skid

Barcodes first, Claude second.

Read a skid label

At the end of a shift you photograph the labels of the skids you finished. The app reads every barcode, works out which value is which, and files the skid under its part. Here’s a made-up label.

PART NO. (P) CX418DMOAA02 DEMO COMPONENTS LTD. MISSISSAUGA, ON SAMPLE LABEL QUANTITY (Q) 240 SUP CODE (V) 000417220 SERIAL (S) 104882317 BATCH/LOT 14.OCT.2026 PACKED 10/14/2026 HU 7 OF 12

Photos stay in your browser. Nothing is uploaded.

Press Read this label. The barcodes are decoded right here, with the same library and rules the app uses.

CH_03

A shift, start to finish

The end of a shift is one page, top to bottom.

End of a shift: pick a station

01Who worked

Who
The site supervisor
Before
Names and times on paper notes.
Now
Everyone’s start and end time, entered once at the top of the page.
Where
End of shift, on the floor

02Skid labels

Who
The site supervisor
Before
Part numbers, quantities and serials copied by hand from each label into that part’s workbook.
Now
Photograph the labels. The barcodes are read in the browser and each skid is filed under its part.
Where
On the floor, with a phone camera

03Each skid

Who
The site supervisor
Before
Filled in by hand, one workbook per part.
Now
Hours, suspect and missing counts, the reject tag, defect photos and who sorted it. A skid that takes two days is picked up again the next morning.
Where
The day page, skid by skid

04Times

Who
The site supervisor
Before
Hours typed again into a daily hours sheet.
Now
Worked out per person and part from the skids, and editable. Anything that doesn’t add up is flagged, never fixed.
Where
The day page, under the skids

05Reports

Who
IQ’s office
Before
Typed again into a monthly sheet.
Now
IQ’s own format: a sorting report per customer group and per part as PDF and Excel, daily and monthly hours, or all of them as one zip.
Where
IQ’s office

Around that page: a reports page that answers “every part we’ve done for this customer and the total rejects” for any date range; a skid board built from the customer’s skid list, oldest first, so shipping can be asked for skids in order; and a company with sites, admins, managers and supervisors, where each company only ever sees its own data.

CH_04

Hours into times

Nobody on the floor writes down when they started each part.

From skid hours to everyone’s times

They know how long a skid took and who was on it. Change the hours or the crew and watch the day lay out. Push the bracket skid to 8 hours, then end Lin’s shift at 9:00, and add Asha to the bracket skid.

Who worked
Skids finished today, in part order
Times per person and part
    CH_05

    The app

    The day page: totals for skids, parts sorted, suspect and missing, and hours; who worked with start and end times; and the first skid, read from its label photo.
    The end of a shift starts here: who worked, then the skids read from their label photos.
    A part without skids with its hours and crew, the times worked out for each person, and a per-person check that everyone’s eight hours are on parts.
    Times are filled in from the skids, and every person’s day is checked against their shift.
    Reports for the day: a sorting report per customer group and per part as PDF or Excel, daily hours and monthly hours.
    One tick per report, or all of them as a zip.
    Reports page: parts and date filters, totals for skids, parts sorted, rejects and hours, and a table per part.
    Any date range, any parts: skids, rejects, reject rate and hours.
    CH_06

    Decisions

    Barcodes first, Claude second

    Barcodes are read in the browser, free and instant. Only when a label is torn or blurred does the photo go to Claude Haiku, which is asked for JSON and told to leave a field empty rather than guess a digit.

    Point out problems, never fix them

    Hours are pay. The app never moves time between people or parts. It splits a skid evenly, keeps everyone inside their shift, and flags what doesn’t add up: past an end time, two parts at once, hours nobody has room for.

    Excel stays Excel

    The office lives in spreadsheets, so the reports do too: IQ’s layout and colours, with live formulas for good parts, totals and suspect rates. Correct one number and everything that depends on it follows.

    Built for more than one site

    Companies, sites and roles are separate in the database, with row-level security on every table. Adding a customer plant is a new site, not a new copy of the app.

    CH_07

    Where it is now

    Status
    Used every shift since late September 2026
    Replaced
    A workbook per part, a daily hours sheet and a monthly one
    Ready for
    More sites, each with its own supervisor and crew
    CH_08

    What I’d change next

    Bring on a second site. The data model is ready for it. What isn’t proven yet is whether another supervisor, on another floor, closes out a shift the way I do.

    Measure the fallback. Barcodes handle most labels, and Claude Haiku only sees the torn or blurred ones. I’d like to count how often that happens and how often a field comes back empty, so the trade-off is a number rather than a feeling.

    CH_09

    Reflection

    Building for my own shift made the first version easy to judge: either the report was done before I left or it wasn’t. The harder call was the rule that the app never moves hours between people. Quietly balancing the numbers would have been simpler. Flagging them means a supervisor has to look, and that is the point: hours are pay.