By industry
AI agents and automation for manufacturing and textiles
On a factory floor the data exists — it is on a production log, a delivery challan, a QC sheet and a supplier’s PDF invoice. What does not exist is agreement between them. We build the layer that reads all four, reconciles them against the ERP, and puts a variance in front of a person instead of letting it surface at month end.
Two weeks to a written build plan · no obligation to build with us
- The reconciliation we are asked for most
- Stock vs booksThe reconciliation we are asked for most
- Nothing posts to the ERP unseen
- Reviewed writesNothing posts to the ERP unseen
- Open-weights models on your own hardware
- On-prem optionOpen-weights models on your own hardware
- To a written build plan, via the audit
- 2 weeksTo a written build plan, via the audit
Where the time goes
If two of these are true, there is something here
Written the way the people doing the work describe it, not the way a process document does.
The books and the floor disagree, and you find out at month end
Consumption is recorded on paper or in a shift spreadsheet, purchases in the ERP, and the physical count once a quarter. The gap between them is discovered long after anyone can remember what caused it.
A parallel spreadsheet is doing the real accounting
The ERP exists, but the number people trust lives in a workbook one person maintains. Every report starts by exporting from the system and then correcting it.
Supplier paperwork is retyped
Invoices and challans arrive as PDFs or photographs. Someone types the lines into the system, matches them to the purchase order by eye, and flags the mismatches from memory.
Quality data never leaves the QC sheet
Defects are recorded in a register or an inspector’s notebook. Nobody can answer "which shift, which machine, which supplier lot" without a manual tally.
What we connect to
The systems you are already running
Manufacturing is where the oldest and the newest systems sit next to each other — a modern ERP, a PLC from 2011 and a book of challans. All three are legitimate inputs.
- ERP and finance
- SAP Business One
- Odoo
- Microsoft Dynamics 365 BC
- Tally
- QuickBooks
- Custom ERP databases
- Plant and shop floor
- MES
- SCADA
- Modbus / OPC-UA
- PLC historians
- Barcode / QR scanners
- MQTT
- Documents and quality
- Supplier PDF invoices
- Delivery challans
- QC inspection sheets
- Excel workbooks
- Scanned registers
What we build here
Five things worth building for manufacturing & textiles
Each one names the systems it touches, because an integration that is not named is an integration nobody has thought about yet.
- 01
Stock and accounts that reconcile themselves
Consumption, receipts and production output read from where they are actually recorded and matched against the ERP daily rather than quarterly. Every difference is presented with both numbers and the documents behind them.
- Touches
- ERP · production logs · stock counts
- What changes
- A variance is a same-day question instead of a month-end argument.
- 02
Supplier document intake
Invoices and challans — PDF, scan or phone photograph — read into line items, matched three ways against the purchase order and the goods-received note, and posted when they agree within tolerance.
- Touches
- Mailbox · ERP purchase module · document store
- What changes
- Retyping supplier paperwork stops; only genuine mismatches reach a person.
- 03
Production reporting without the shift spreadsheet
Output, downtime, rejects and consumption captured at source — scanner, MES or a simple shift form — and turned into the report management currently builds by hand every morning.
- Touches
- MES / scanners · ERP · reporting
- What changes
- The morning report exists before the morning meeting.
- 04
Visual quality inspection
Camera-based detection of the defects your inspectors already look for, running on the line with confidence scores and a saved image for every call. Borderline cases are routed to a human rather than decided.
- Touches
- Line cameras · QC records · ERP lot data
- What changes
- Defect data becomes queryable by shift, machine and supplier lot.
- 05
Maintenance and downtime intelligence
Machine logs, breakdown records and maintenance history pulled together so a recurring fault is visible as a pattern rather than as a series of separate incidents.
- Touches
- PLC / historian · maintenance log · ERP asset records
- What changes
- Downtime causes stop being rediscovered every quarter.
How it is put together
How a supplier invoice becomes a posted entry
This is the flow most factories start with, because it is the one where the manual work is most obvious and the rules are clearest.
- 01
Capture
Mailbox, scanner or upload. The original file is stored untouched and everything downstream references it, so a dispute with a supplier can always be settled against what they actually sent.
- 02
Extract
Line items, quantities, rates, taxes and the PO reference pulled out with a confidence score per field. Low-confidence fields are flagged rather than guessed — an invoice with a smudged quantity is not a place for optimism.
- 03
Match
Three-way match against the purchase order and the goods-received note, with your unit conversions and rounding rules applied. Partial deliveries and split invoices are treated as normal cases, not errors.
- 04
Tolerance
Your tolerance — by value, by percentage, by supplier — decides what may post unattended. Everything outside it goes to the review queue with both sets of numbers side by side.
- 05
Post and log
A scoped ERP credential that can create the entry and nothing else, with an idempotency key on the invoice number so a retry cannot double-post. The decision, the confidence scores and the approver are all logged.
Your data
What we will and will not do with it
Stated plainly, and without claiming a certification we do not hold. If a requirement here is harder than what is written, say so on the call and we will tell you whether it is buildable.
It can run without sending anything to a model vendor
Factories often cannot send commercial terms to a third-party API, and several of ours will not. Document extraction and classification can run on open-weights models on hardware in your own plant or private cloud. It costs more to set up and it removes the question entirely.
We do not write to control systems
We read from PLCs, historians and MES. We do not write setpoints or issue machine commands. Process safety is not a place for a language model, and no commercial argument changes that.
Financial postings stay inside your tolerance rules
What may post unattended is defined by you as a number, not inferred by a model. A supplier or a value outside that range always stops for a person, however confident the extraction was.
Proof
We have taken the accounting off the spreadsheet before
Base Papers’ managing director wanted the accounting automated, because the numbers people trusted were living in a workbook beside the system. We replaced the parallel spreadsheet with a system the stock and the accounts actually live in — which is the same problem in this sector, whatever the ERP is called.
- For stock and accounts
- Single sourceFor stock and accounts
- Floor to ledger
- Real-time syncFloor to ledger
Technology
What this gets built with
We pick the boring option unless there is a reason not to — the framework is the part most likely to be abandoned before your system is.
- Extraction
- Document AI
- OCR
- Layout parsing
- Structured output
- Confidence scoring
- Vision
- YOLO
- OpenCV
- PyTorch
- ONNX Runtime
- Edge inference
- Integration
- SAP B1 Service Layer
- Odoo XML-RPC
- Modbus / OPC-UA
- MQTT
- Python
- Platform
- PostgreSQL
- Celery
- Docker
- On-prem deployment
- Audit log
How the engagement runs
Five phases, each with something you can hold
Every phase ends in a named deliverable. You can stop after any of them and keep what has been built.
- 01Week 1–2
Sit with the work
We watch the process as it is actually done, not as the process document says. The output is a map of every step, which system it touches and how long it takes.
DeliverableProcess map and a shortlist of what is worth automating - 02Week 2–3
Design the access layer
Which systems get read, which get written to, what scope each credential carries, and which decisions stop for a person. Agreed in writing before anything is built.
DeliverableArchitecture and permission model - 03Week 3–8
Build against your real data
Built in your sandbox against real records, not a demo dataset. The write path goes through a review queue from the first day, so nothing lands in a live system unseen.
DeliverableRunning system in staging, with the review queue - 04Week 8–10
Run it beside the humans
It runs in parallel with the people doing the job. We compare the two outputs, fix what differs, and only then start letting the routine path through unattended.
DeliverableAgreement rate report and a go-live threshold - 05Ongoing
Hand over and watch
Source code in your repository, deployment in your accounts, cost and failure dashboards you can read without us. Support if you want it, not because you are locked in.
DeliverableRepository, runbook and monitoring
Why Devs Core
Three reasons that are checkable
We run two of our own products
Cadenza and Budge are ours. What an agent costs per run, where it fails and which decisions need a person came from operating them, not from a vendor deck.
The write path is reviewed by default
Anything that changes a record in a system of record goes through a queue a person can see. You choose what graduates to unattended, and when.
You own the code
Source in your repository, infrastructure in your accounts, model keys in your name. There is no runtime of ours you have to keep paying for.
Questions
Manufacturing & Textiles — the questions we get asked
Our ERP is old and has no proper API. Is this still possible?
Usually yes. Older ERPs almost always expose a database, a scheduled export or an import format, and we have worked through all three. What we will not do is screen-scrape a UI and call it an integration — if that is the only option we will say so before you commit to a build.
Can this run inside our factory network, with no internet?
Yes, with open-weights models running on your own hardware. Expect to provide a GPU machine and to accept slightly lower extraction accuracy than the largest hosted models. Several manufacturers consider that trade obviously worth making.
Will an AI make purchasing or production decisions?
No. It reads, matches, reconciles and proposes. Purchase approvals, production planning changes and anything touching a control system stay with your people. The value here is in the reading and matching, which is where the hours go.
How accurate is invoice extraction on poor-quality scans?
Good on clean documents, variable on phone photographs of carbon copies — which is why every field carries a confidence score and low-confidence fields are held rather than posted. The honest measure is not accuracy on a test set; it is how many invoices a person still has to touch after a month, and we report that number during the parallel run.
We make textiles. Does that change anything?
Mostly in the units and the lot logic — shade lots, roll lengths, GSM, seconds and wastage rules are where generic systems get it wrong. Those become explicit rules rather than assumptions, and they are the part we ask the most questions about during discovery.
What about quality inspection — can a camera really replace an inspector?
It should not try. It catches the defect classes it was trained on, at speed, consistently, and routes anything borderline to the inspector. The gain is that every call is recorded with an image, so quality data becomes something you can analyse rather than a tally at the end of a shift.
Related
What manufacturing & textiles buys most often
- Read more
Process Automation
Replace the manual steps between two systems — the re-keying, the copy-paste, the spreadsheet that reconciles what neither system knows.
From $4,500 - Read more
Computer Vision Development
Detection, inspection and counting from camera feeds, running on the floor rather than in a research notebook.
From $9,000 - Read more
AI Integration Services
The access layer between the software you already run and any AI model — so an agent can reach your ERP, CRM or helpdesk without being handed the keys.
From $8,000 - Read more
Custom Software Development
The operational system underneath — the one the packaged software did not cover. React and Django, built to be handed over.
From $14,000
Tell us which process is costing you.
Thirty minutes, no preparation, no deck. You describe what keeps going wrong and we tell you whether AI is the answer — including when it is not.
Prefer email? contact@devs-core.com