MANUAL GENERATOR
AI workflow for installation manuals
Built.
Redesigned.
CONSTRAINTS
VARIABLE PROJECT INPUTS
Each project arrives differently: notes, images, dimensions, material lists and remembered details. The system had to work with uneven information instead of assuming a fixed form.
MISSING INFORMATION
If something important was unknown, the system could not silently invent it. Missing information became an explicit project state that stopped the draft and asked for clarification.
HUMAN REVIEW
The manual is used by people installing physical artworks. Generated content therefore had to remain reviewable, editable and under human control before final output.
DECISIONS
NOTION AS THE V1 INTERFACE
V1 hid the AI behind a familiar project table. The assistant entered project information in Notion, changed the status to Generate Draft, and the workflow ran behind the scenes.
REBUILD THE WORKFLOW
V1 worked, but the n8n system became too dependent on its builder understanding the plumbing. V2 replaced that brittle automation with a simpler, conversational workflow.
IMAGES MUST EXPLAIN SOMETHING
V2 stopped treating images as decoration. It could select, crop or annotate a useful detail, ask for another angle, or leave a page without an image when none added information.
IMPACT
2–3 DAYS → 1–2 HOURS
The old process could stretch across two or three working days. The redesigned workflow brings the same job down to roughly one or two hours including review.
LIGHT GRID TEST
The configured test produced a seven page installation manual in about ten minutes, then went through three review rounds before the final version was approved.
HUMAN IN THE LOOP
The final system proposes the Installation Record, page plan, image use and draft manual. A person still reviews the content, resolves uncertainty and approves the PDF.
Workflow design / Product design / System architecture / Prompt design / Prototyping / Testing / AI implementation