Skip to main content
Quinn

Working with Quinn

The four things Quinn does — documents, expert knowledge, structuring, querying — plus how to write prompts.

Written By Tommy Giesbrecht

Last updated About 1 month ago

Article goal: After reading this article, you will know which tasks Quinn takes on, which input leads to which result, and how to phrase a prompt so the result is fast, short, and easy to check.

Relevant for the Editor and Administrator roles. Quinn is currently in beta.

Quinn is currently optimised for processing documents and transferring them into the findIQ method. Further use cases are planned — we look forward to your feedback.

1. Reading documents

  • Input: operating manuals, troubleshooting tables, tickets, spreadsheets (PDF, Word, Excel).

  • What Quinn does: reads the fault information they contain, maps it to the findIQ model — symptoms, causes, routines, labels — and creates the matching template or extends an existing one.

  • Result: entries in findIQ with the same structure as manually created ones. The content ends up in the data model, not in a file store.

2. Recording expert knowledge

  • Input: a conversation in the chat with someone who knows the machine.

  • What Quinn does: asks about missing details, classifies the answers as symptoms, causes, and routines, and enters them.

  • Result: field knowledge recorded in findIQ without the person answering having to know the findIQ method. The capture is asynchronous — not tied to a meeting or a location.

3. Structuring existing knowledge

  • Input: a template in findIQ with symptoms and causes, plus a category scheme — either one you specify or one Quinn proposes.

  • What Quinn does: analyses the symptoms and causes and labels them according to that scheme, for example as a labelled heatmap.

  • Result: a structure that troubleshooting can follow. Large knowledge bases are processed in one pass instead of entry by entry.

4. Querying and summarising the data

  • Input: a question about your organisation’s data — machines, templates, routines, logbook.

  • What Quinn does: searches across the hierarchy within your permissions and summarises what it finds.

  • Result: an answer without clicking through nested levels. Filtering, sorting, and browsing remain faster in the interface.

Writing prompts

A prompt works better the more precisely it says where Quinn should work. What findIQ already knows does not need explaining.

📝 A complete prompt

"Work in the existing template Bottle filler. From the attached operating manual, use only the troubleshooting table T-01 to T-09 and, from it, the faults of assembly AS-3. Add the symptoms and causes to the heatmap and do not create a new template."

Four pieces of information are in there: the target object, the section, the action, and a boundary. The eight points below go through them.

1. Name the section, not the whole document

An anchor keeps the data volume small: plant, assembly, section, table range. The result arrives faster and is easier to check.

  • Instead of: "Analyse this operating manual."

  • Better: "Use only the cover sheet and, from the troubleshooting table (T-01 to T-09), the faults of assembly AS-3."

2. Don’t explain what findIQ already knows

The setup and logic of the heatmap are predefined. Describing them makes the prompt longer, not better.

  • Instead of: "Build a heatmap with symptoms as rows, causes as columns, and probabilities based on frequency."

  • Better: "Add the tickets for component AS-6 to the heatmap."

3. Name the object to work in

Without a target, Quinn may create something new instead of extending what exists.

  • Example: "Work in the existing template Bottle filler and do not create a new one."

4. Work in steps

Add, label, check — one step at a time. If a step doesn’t fit, you don’t repeat everything.

  • Instead of one prompt: "Read in the manual, label everything by assembly, and check the probabilities."

  • Better, three in sequence: "Create the faults of assembly AS-3." → "Label the new symptoms by assembly." → "Show me the symptoms that still have no cause."

5. Specify, or let Quinn suggest

Specifications give reproducible results. Open questions make use of Quinn’s view of the data.

  • Reproducible: "Create a label Pressure system and label all symptoms and causes belonging to the pressure system (CO2 pressure)."

  • Collaborative: "I want to label symptoms and causes by fault type. Which labels do you suggest?"

6. Use the terms of the application

Template, label, routine, symptom, cause, logbook — each means something specific in findIQ.

  • Example: calling an Excel file "logbook" while the logbook is a feature invites misunderstandings.

7. Use the interface where it is faster

Quinn is strong at evaluating, structuring, and creating. Browsing is more direct in the interface.

  • In the interface: finding a template, filtering the heatmap by label, correcting a single entry.

  • With Quinn: creating 60 faults from a table, labelling every symptom of the pressure system, summarising what is in the data.

8. Decide in advance what the result should be

If you know the expected answer, deviations are obvious straight away.

  • Example: "Table T-04 should produce four faults, because four belong to AS-3." If five come back, you know where to look.

  • Example: before labelling, note that the pressure system should end up with roughly a dozen symptoms. If two come back, the category was drawn too narrowly.

Quinn follows the findIQ method and asks when information is missing instead of filling the gap. A precise prompt makes processing more efficient, but it is not a prerequisite.

Read next