PrintMind

PrintMind is my bachelor thesis: an AI-assisted production planning system for a mid-sized print shop, designed and built end-to-end — from field research on the shop floor to a working prototype. Print production planning still happens on physical peg boards and in spreadsheets. Distributing print jobs across machines and shifts by hand is slow, error-prone and has to respect a web of constraints: machine capabilities, print approvals, coating blocks, spot colours, delivery dates. PrintMind digitises this process and adds an AI layer that does the heavy lifting — while keeping the final decision firmly with the people who carry the responsibility.

• 3 roles with their own workflows: project management, prepress, production lead
• 1 real print shop as research and testing partner

The production lead of a print shop juggles dozens of jobs across machines, shifts and finishing steps — under constant time pressure and with rush jobs cutting the line daily. The existing tools don't help: the ERP system knows the orders but not the plan, the peg board knows the plan but nothing else, and none of it is visible to project management or prepress. The result: planning knowledge lives in one person's head, changes are laborious, and every phone call about a job's status interrupts production.

The project was grounded in a real print shop: I spent time on the shop floor, interviewed all three roles — project management, prepress and production lead — and mapped how a print job actually travels through the company. Usertests happened on site, with iterations built directly on the feedback.

1. 01

   The peg board works — its spatial logic is deeply learned. A digital tool must preserve it, not replace it with abstract lists

2. 02

   Trust is the bottleneck for AI planning: a proposal that cannot be inspected and corrected will simply not be used

3. 03

   Three roles need three different views on the same data — a single generic interface serves nobody

4. 04

   Rush jobs are the norm, not the exception: replanning must take seconds, not minutes

5. 05

   Corrections are knowledge: every manual fix of an AI proposal is training signal for the next one

PrintMind is the peg board — digital. A weekly plan shows every machine with its shifts; jobs are placed via drag & drop, can be split across night shifts and moved across day boundaries, exactly like the physical original. On top sits the AI planning: imported ERP jobs are pre-sorted, and on request Claude generates a complete weekly schedule that respects machines, deadlines, coating blocks and rush jobs. The proposal lands in a review state — the production lead inspects, corrects and only then releases it. Corrections flow back into a feedback loop that improves future proposals. Each role gets its own view: a status board for project management, a job pipeline for prepress, dashboard and planning for the production lead.

• Digital peg board: weekly plan with drag & drop, cross-day moves and night-shift splitting
• AI proposal with approval gate: Claude plans the week, the production lead decides — never automatic
• Feedback loop: manual corrections become context for future planning runs
• ERP integration: jobs are imported from the existing system via XML — no double data entry
• Role-based views: project management, prepress and production lead each see what they need

The working prototype was tested on site with all three roles at the partner print shop, and iterations were built directly from their feedback. The core bet of the thesis held up in testing: AI planning is accepted when it behaves like a competent colleague making a suggestion — transparent, correctable, and never acting past the human. Beyond the thesis itself, PrintMind is the project where my two directions finally converge: interface design that respects how people actually work, and an AI system architecture — prompt design, guardrails, feedback loops — built around trust.