Manufacturing
AI for manufacturers
and the small factory that runs on memory
Most small factories keep twenty years of knowledge in four people's heads. My job is to move that knowledge into a system the business owns.
A small factory usually runs on memory. The supervisor knows which machine drifts, the storekeeper knows which supplier delivers short, and the owner holds the rest. None of it is written down, so none of it can be asked a question.
I build the writing-down part. Shift handover, job cards, breakdown notes, dispatch and attendance all go into one set of records, mostly from a voice note or a short chat in the language your supervisor already speaks. The records sit on a machine in your own plant office. There is no monthly software subscription, and your accounts software can be connected to it.
The shift handover writes itself
The outgoing supervisor sends a voice note. The Agent turns it into the handover note, files it against the machine and the job, and the next shift reads it before starting.
Job cards without the clipboard
Quantity made, rejected and reworked, against the job and the operator. Entered once on a phone, visible to the owner without anybody preparing a report.
Attendance and leave off paper
The fingerprint machine you already own feeds the muster, the overtime and the leave balance. I have built this for a forty-person studio and for my own team of eighteen.
Dispatch paperwork, filled once
The dispatch record, the packing detail and the documents that travel with the lorry, filled from the order instead of typed again at the gate.
Ask the records a question
What did we make last month on the second line, which supplier is late, which customer has complained twice. Asked in plain words, answered from your own data.
It stays inside the plant
Drawings, costings, supplier prices and salary sheets are the things nobody wants to paste into an outside chat window. On your own machine they do not leave.
The day the supervisor who knows the machine is on leave
Nothing is wrong with running on people. It is how most Indian MSMEs became good at what they do. The problem arrives on the day the supervisor who knows the machine is on leave, or the storekeeper retires, or the owner's son takes over a plant he has only seen from the office. Every answer has to be asked of a person, and the person is on the shop floor.
The second problem is the retyping. The same order is written on a pad, keyed into a sheet, read out on the phone, typed into accounts and copied into a dispatch register. Each copy is a chance to be wrong, and the clerk who holds all five copies in his head is a single point of failure nobody has planned around.
The paperwork a plant cannot avoid
A factory carries an unusual amount of compulsory writing. Purchase orders arriving by email and WhatsApp in a different format from every distributor. Job cards. Rejection and rework notes. Material issue slips. Dispatch documents and the e-way paperwork that travels with the goods. Machine breakdown and maintenance logs. The file the quality auditor asks for when the ISO visit is due, which is usually assembled in a panic out of memory.
None of this needs AI to decide anything. It needs the facts captured once, where they happen, and written into the right record afterwards. That is what an Agent is good at, and it is also the work your people least want to do.
Accounts software, including Tally, can be connected so these records are not typed twice. I have not built a Tally connection for a client yet, so it is scoped and tested during the audit before anybody promises it.
The March count, the festival rush and the shutdown
A plant's year has a shape. Orders climb before the festival season and the distributors want everything at once. The financial year closes in March with a stock count that eats a week and produces numbers nobody fully trusts. The annual shutdown is when maintenance actually happens, and the notes from it are usually lost by the time the same machine fails again.
A system that only works in a calm month is useless here. I build for the busy end: capture that works standing next to a machine, records complete enough for the March count to come out of the system instead of out of a notebook, and a maintenance history that is still readable next year.
Where to start in a small factory
Start with one line or one department, and with the record that is costing you the most arguments. For most plants that is either the shift handover or the job card, because both decide what you can bill and what you have to make again. Attendance is usually the easiest win and the one the staff notice.
- One line, one shift, one record to begin with
- Shift handover from a voice note
- Job card entered on a phone at the machine
- Attendance and leave from the device you already own
- A dispatch record that fills itself from the order
- One screen the owner reads without asking for a report
What I do not do for manufacturers
I do not fit sensors to machines, read vibration data or sell predictive maintenance. I do not do machine vision for inspection on the line. If somebody has told you AI will warn you before a bearing fails, that is a different trade with a different supplier, and it needs hardware on the machine before any software matters.
What I do is the office and the paperwork around the plant: reports from a voice note, attendance, leave and daily work, and a private AI server that keeps costings and drawings in the building. For a company of forty to a hundred people, this page is closest to your shape. I am also rebuilding the website for Sunstar Graphics, a screen printing machinery distributor, which is in production now.
Built, not described
Questions people ask
Can AI help a small manufacturing business in India?
Yes, in the office more than on the machine. The useful work is capturing the records a plant already has to keep, shift handover, job cards, dispatch and attendance, then answering questions from them. Sensors and inspection cameras are a different trade that I do not sell.
Will my supervisors actually use it?
They use it when it is the thing they already do. A voice note at the end of a shift, in the language they speak, is not a new habit. Nobody in my own studio of eighteen sat through a training session, and that is the bar I build to.
Can it write into our accounts software or Tally?
Accounts software including Tally can be connected, so a record is entered once instead of twice. I have not built a Tally connection for a client yet. It is scoped and tested during the audit rather than promised before it.
We have no IT person. Who runs this?
The machine sits in your plant office and restarts itself after a power cut. I hand over the logins, the code and a written explanation. Nothing I build needs an IT department to keep running.
What does it cost to digitise an MSME plant?
Two plants of the same size can need very different builds, so there is no rate card. After the audit you get a quote naming the machine, the records being built and the handover. The build is paid for once, and nothing is charged monthly per worker.
Will our drawings and costings go to an outside AI company?
Not if the model runs in your own plant. That is the point of an on-premise setup. Your drawings, prices and salary sheets stay on the machine you bought, and I document who can reach them.
Is this an ERP?
No. An ERP tries to be every record in one product you license. I build the few records you actually argue about, on hardware you own, connected to what you already use. If you want a full ERP later, these are the records it would need anyway.
Next step