An AI maintenance assistant for pharmaceutical manufacturing

A technician describes the problem out loud, in English or Urdu. The assistant asks the right questions, suggests the most likely causes in the safest order to check, and writes up the maintenance ticket.

TP-01 · Tablet compression machine: Try: "TP-01 is capping, the top of the tablet is coming off", or "alarm E-12 is showing".

BP-02 · Blister packing machine: Try: "Blisters are failing the leak test", or "A-07 low air pressure alarm".

FBD-03 · Fluid bed dryer: Try: "Drying is taking much longer than usual", or "F-01 earthing alarm".

AC-04 · Steam autoclave: Try: "The autoclave is not reaching temperature", or "the door won't open".

1. Describe the problem: Say the machine number and what you see, hear or measure, in English or Urdu.

2. It asks what matters: Alarm code, when it started, whether a batch is running, and what you already tried.

3. Safest check first: It gives the two or three most likely causes and walks through checks one at a time.

4. Ticket written: It reads back a ticket: machine, problem, checks and results, likely cause, next action, batch impact and escalation.

Lockout before hands-on: Every hands-on check starts with stopping the machine and lockout/tagout under your SOP.

Never bypasses safety: It refuses any request to bypass, jumper or disable an interlock, guard, alarm or sensor.

Stops for danger: Injury, fire, a steam or gas leak, or an earthing alarm: stop, get clear and follow the site emergency procedure.

GMP aware: If a batch may be affected, it tells the technician to inform the supervisor and QA and record a deviation or change.

Stays inside validated ranges: It never suggests a setting outside the batch record or SOP range, and never says a batch can be released.

Never guesses: If the answer isn't in the machine notes, it says so and escalates to the maintenance engineer.

Your machines mapped: Each machine set up from its own manual, alarm code list and your SOPs, and reviewed by your engineering and QA leads.

Tickets in your system: Tickets sent to your maintenance system or email, so breakdown history builds up per machine.

Learning from history: Past breakdowns and fixes for each machine added to what it knows, so common faults get faster.

On the floor: A phone number or a tablet at each line, with Urdu and English, and a noisy-room headset.