A parts manufacturer's paint line foreman retires in 18 months, and 30 years of his practice has never been written down. Corporate's answer was to order a 200-page knowledge transfer document. Nobody standing in front of a pressure alarm at 2 a.m. reads 200 pages from the top, so the on-call tech calls the foreman instead. In 18 months there will be nobody to call.
The production engineering manager who owns the line has 15 years of scanned handwritten maintenance logs, an Excel export of MES comments, 22 equipment manual PDFs, and a 40-minute interview he records on the line each week. This case covers how that material goes into Consilience, and what gets asked of it afterwards.
The material: maintenance logs, MES records, and interviews
Most of this line's knowledge is already written down. It is split across A4 scans, spreadsheet cells, manual PDFs, and the foreman's memory.
- 15 years of scanned handwritten maintenance logs. 180 PDFs, roughly 4,200 pages, in ballpoint handwriting mixed with abbreviations and sketches
- One Excel export of MES equipment comments. About 30,000 rows going back to 2018, each cell holding a single line such as "P-103 hunting again, drained trap"
- 22 equipment manual PDFs for paint booths, transfer pumps, and regulators, straight from the manufacturers
- Line interview recordings. One 40-minute m4a a week, the foreman talking in front of the machine
- Three names for the same machine. The logs say "No. 3 Booth," the MES says "P-103," and the manuals and work standards say "Paint-3"
How it works today
Corporate asked for a 200-page knowledge transfer document. The foreman writes it in Word after his shift, in table of contents order, and is about halfway through. The finished pages are printed in a binder in the line office cabinet.
The binder fails at the moment it is needed. Finding "No. 3 Booth" in the contents does not surface the 30,000 rows of MES comments filed under "P-103" over three years, because nothing ties those two names to the same machine. The material exists and the search cannot reach it.
How he sets it up
Setup costs the manager one day in the first week. After that it is one recording a week.
Building the paint line workspace
Load 15 years in one run
Open "Extract sources" from the left sidebar (
Cmd+Shift+I), drag in the folder of scanned logs, the MES spreadsheet, and the 22 manual PDFs, then press "Import". The handwritten scans go through OCR to get the letters out. Progress runs through five stages: "Reading your files" → "Getting to know your topics" → "Extracting the knowledge graph" → "Connecting related items" → "Organizing into themes". The "Getting to know your topics" stage picks up this line's vocabulary, including hunting, trap drain, and booth differential pressure, and extraction uses it from there on. The original PDFs stay exactly where they were, underattachments/.Confirm the three names as one machine
At "Connecting related items", a proposal comes up to merge "No. 3 Booth," "P-103," and "Paint-3" into one machine. Nothing is applied automatically. The manager has to confirm the three before any name returns the same material. If he confirms a wrong one, it can be undone.
The foreman just talks
Each week's 40-minute m4a from the line goes into the same "Extract sources" wizard. Audio is transcribed into a note, and that note runs the same five stages into the graph. The fixes and judgment calls the foreman spoke get attached to the machine, sourced to that interview note. Nowhere in this flow does the foreman write a Word document.
What he asks
"booth 3 pressure hunting, past cases on humid days"
→ Five cases come back as a list. Each one carries the symptom and
the fix from the time, plus a source note link (maintenance-log_2019-08-03).
"pull together the recurring symptoms and fixes for P-103"
→ Grouped into three symptoms: pressure hunting, nozzle clogging,
transfer pump pulsation noise. Under each, the fix and source note
links (maintenance-log_2021-05-14, MES comment).
Asking for "No. 3 Booth" returns the same answer.
"what's the current regulator replacement interval on booth 3"
→ Three months. The basis is what the foreman said in a 2023 interview note.
The 6-month interval from the 2016 log is shown alongside it as the old one.Two things have to be true for that answer to come back. First, the three names are confirmed as one machine, so the logs' "No. 3 Booth" and the MES's "P-103" pile up in the same place. Second, clicking the note link on the answer opens the note that came out of the 2019 ballpoint scan, so the on-call tech can check the original before he trusts the answer.
Until the manager confirms them, the old interval and the new one sit in their own notes as they are. Which of 6 months and 3 months is current only shows up in the answer after he checks both notes and resolves it.
What changes
| Before | Now |
|---|---|
| At 2 a.m. you hunt through a 200-page binder's contents, then call the foreman | Type the equipment name and past cases come back as a list, each with the log note it's based on |
| Search "No. 3 Booth" and the 30,000 rows of "P-103" comments in the MES don't come up | The three names are confirmed as one machine, so any of them returns the same answer |
| The foreman writes 30 years into Word after his shift | The foreman talks on the line; the weekly 40-minute recording becomes a note attached to the machine |
| 6 months from 2016 and 3 months from 2023 sit in separate documents, and the documents alone don't tell you which is current | The current interval comes first in the answer, and the old one is shown with the note it came from |
For 18 months, the manager's repeat work is dropping one recording a week into "Extract sources" and confirming the name merges and interval changes that come up. Setup cost one day in the first week; after that it's a few minutes a week.
Try this workflow with your documents
Learn the steps behind this workflow, then try them with your own sources.
Open the lesson ↗