AI · Built into the Textile Mill ERP

ERP Assistant — Ask the System by Text or Voice

Staff ask in plain words — typed or spoken — things like "what is left to print on this lot?" and get the same figure the report shows. It runs reports, explains screens and opens forms pre-filled for a person to save. It only shows what the asking user is allowed to see, never writes data itself, and runs fully on the factory server, so business data never leaves the building.

The problem

A factory system has dozens of screens and reports. Most staff use two or three of them and are lost in the rest, so simple questions — how much of this lot is left to print, where is this order — turn into phone calls to the one person who knows the system.

How it works

Staff type or speak a question in plain words. The assistant works out what was asked, checks what that person is allowed to see, and only then picks the best answer: a report, an explanation of a screen, or a form opened with the values already filled in for a person to save.

  • It calls the same reports the screens use, so its number always matches the report
  • It forwards the user's own login, so it can never show more than that user could open themselves
  • It never writes to the database — a person always presses Save
  • When it is not sure, it offers the closest options instead of guessing

Built for a real factory server

The assistant does not send company data to an outside AI service. It runs fully on the factory's own modest server, answers instantly, and works when the internet is down. It is designed so a larger language model can be plugged in later if the hardware allows.

Voice

A microphone button sits beside the question box. The words it heard are shown in the conversation, so a mishearing is seen at once instead of turning into the wrong report.

Technology

Node.jsCustom NLU engineVoice inputPermission-aware tools

More work

Other projects

AIFactory production line

AI Machine-Safety Interlock

Cameras watch a production line and, when a person is somewhere unsafe, the machine near them is stopped through the PLC. One batched detection pass covers 15 cameras and controls two machines independently. Mask zones separate "outside the plant wall" from "allowed to stand here", and the design is fail-safe: when in doubt, stop.

PythonYOLOv8OpenCVRTSP camerasModbus · Delta PLC
ERPTextile processing mill · Live, in daily use

Textile Mill ERP

A complete production management system that follows fabric from greige receipt through printing, dyeing and post-processing to packing and dispatch, with lot-level traceability. It replaced paper, spreadsheets and a single-user Access database. I built and maintain all of it alone: 45 screens, a 79-table database, 31 API modules, tablet layouts for the production floor and an Android companion app.

React 19ViteNode.jsExpressMySQLSheetJSAndroid
AutomationPart of the Textile Mill ERP

Reporting, Backup & Audit Automation

Production summaries email themselves to management twice a day, with WhatsApp supported as a second channel. Nightly database backups are verified, compressed, uploaded off-site to Google Drive and pruned after 30 days. A column-level audit trail records who changed which value and when, so a disputed quantity always has an answer.

node-cronNodemailermysqldumprcloneTask Scheduler

Contact

Tell me the problem. I will show you the fix.

Have a similar problem in your business? Describe it and I will reply with how I would solve it.