Identify
Understand exactly where manual MTR review broke down and why cross-checks weren't catching it
The situation
Graphite India Limited manufactures graphite electrodes and other carbon and graphite products for steel producers and process industries worldwide. Every consignment ships against a certified material specification, and every batch of incoming raw material has to be traceable back to its source heat and verified against the engineering bill of materials (EBOM) before it enters production.
On the compliance floor, the team had already lived with:
- Manual, document-by-document cross-checking of MTRs against EBOM tolerances
- No structured link between a shipment and the documents that had cleared it
- MTRs, delivery notes, and invoices arriving in English, Japanese, Korean, and Chinese, with no shared layout
- Grade mismatches and out-of-tolerance compositions occasionally slipping through undetected
- Errors that surfaced only after material had already reached the production floor
As supplier volume grew and document formats multiplied, the cost of those errors — scrap, rework, after-the-fact compliance investigations — kept accumulating. The intent to fix it properly, though, was never in question.
The challenge
The challenge wasn't just reading documents faster — it was making it structurally impossible for a non-compliant material to leave the review queue undetected, in an environment where MTRs, delivery notes, and invoices arrive from dozens of suppliers, in four different languages, with no shared layout between them.
Our approach
We were brought in to design a system that would read and verify documents the way a trained compliance officer does — not the way a generic document-scanning tool assumes a document should look.
Our objective was to:
Design
Build an OCR and LLM extraction pipeline that reads any supplier's document format and language
Gate
Make it structurally impossible for a non-compliant shipment to reach production unresolved
Scale
Support a multilingual, multi-supplier document flow with a full audit trail behind every decision
What we did
1
Built a proprietary OCR and in-house LLM extraction pipeline
Every MTR, delivery note, invoice, and EBOM is processed through a proprietary OCR pipeline paired with an in-house large language model, regardless of which supplier issued it or what template they used.
Built to handle real supplier documents:
- Scanned PDFs, faxed reports, and supplier-branded templates with no shared structure
- Documents in English, Japanese, Korean, and Chinese, sometimes mixed on the same page
- No pre-configured template required per supplier
The result: every field — grade, heat number, chemical composition — normalised into one consistent record regardless of source language or layout.
2
Built an automatic EBOM and purchase order cross-check
Once an MTR is extracted, the system automatically compares grade, chemical composition, and mechanical properties against the tolerance ranges defined in the EBOM and purchase order.
One pipeline, two checks:
- Material compliance — grade, composition, and heat number checked against EBOM spec
- Delivery and invoice reconciliation — quantities and line items checked against the PO
- Both checks write to the same structured review queue
3
Added a mandatory compliance review gate
No shipment can be marked ready for production until every flag raised against it has been resolved — either cleared by a compliance officer with a logged justification, or rejected outright.
What the gate covers:
- Extracted values shown side-by-side with the source document and the failed requirement
- A flag blocks release — no silent override without a logged decision
- Addresses the most common cause of non-compliant material reaching production
4
Deployed on-premises with a full audit trail
The platform runs on Graphite India's own infrastructure, accessible from desktop browsers in the compliance office and from tablets and PDAs already in use on the receiving dock.
Supporting details that mattered on the floor:
- Four-language extraction (English, Japanese, Korean, Chinese) normalised into one consistent record
- Compliance dashboard with flag rates by supplier and outstanding reviews
- Every extraction, cross-check, flag, and reviewer decision logged with timestamp and user identity
The compliance team now has a complete, queryable record linking every component back to the MTR and heat number that cleared it.
How it works
From document capture to a released, fully audited shipment — every stage recorded, one hard gate before anything reaches production.
1
Document capture
MTR, delivery note, and invoice captured on tablet or PDA and linked to the purchase order.
Web app · tablets & PDAs2
OCR + in-house LLM extraction
Structured fields extracted regardless of supplier layout or source language.
EN · JA · KO · ZH3
EBOM & PO cross-check
Grade, composition, and quantities compared automatically against tolerance and order.
Rule engine4
Compliance review gate
Every flag must be cleared or rejected before a shipment can be marked ready.
Hard block on open flags5
Audit trail & compliance dashboard
Every extraction, check, and decision is logged and visible to the compliance team in real time.
Full heat-level traceability“
We used to catch compliance issues after the fact. Now the system catches them before material ever reaches the floor — in whatever language the paperwork arrives in.
Graphite India
“
I used to spend most of a shift matching numbers across three documents by hand. Now I see compliance, quantity, and invoice status in one screen before the material even leaves the dock.
Graphite India
0
Non-compliant releases
Since rule-engine rollout
100%
Material traceability
Every component to its source heat
4
Languages supported
English, Japanese, Korean, Chinese
~85%
Faster document review
Time per shipment, pre- to post-launch
The outcome
What changed after rollout:
- OCR + LLM extraction and automatic cross-check against EBOM and spec, in four languages
- Full audit trail linking every component to its source MTR, heat number, and reviewer
- Delivery note, invoice, and purchase order reconciled in the same review workflow
- Compliance dashboard with flag rates by supplier and outstanding reviews
- Non-compliant material flagged and blocked before it ever reaches production
- Heat-level traceability let the compliance team scope a supplier quality issue to the exact affected components within minutes
Frequently asked questions (FAQs)
1. What is Document AI for material compliance?
It's a system that reads Mill Test Reports, delivery notes, invoices, and EBOMs automatically, extracts the values that matter for compliance, and cross-checks them against engineering and purchasing requirements before material is released to production.
2. Why is manual MTR review prone to error?
Every supplier issues MTRs in a different layout and, often, a different language — a compliance officer has to locate the same values in a different place every time, which makes grade mismatches and out-of-tolerance compositions easy to miss under volume and time pressure.
3. How does the platform read documents in different languages?
A proprietary OCR pipeline handles the raw text and layout, and an in-house LLM recognises the same underlying field — grade, heat number, composition — across English, Japanese, Korean, and Chinese, normalising it into one consistent record regardless of source language.
4. What is the EBOM cross-check and why does it matter?
It's an automatic comparison of a shipment's extracted grade, chemical composition, and mechanical properties against the tolerance ranges defined in the engineering bill of materials — any deviation is flagged before the shipment is accepted into inventory.
5. How are delivery notes and invoices reconciled?
The same extraction pipeline reads delivery notes and invoices and matches quantities and line items against the purchase order, surfacing short shipments or discrepancies in the same review queue as material non-compliance.
6. What happens when a shipment is flagged non-compliant?
It routes into a review queue where a compliance officer sees the extracted values next to the source document and the requirement it failed, then either clears the flag with justification or rejects the shipment — nothing flagged moves forward on its own.
7. Can this integrate with existing receiving-dock hardware?
Yes. The platform runs as a web app accessible from desktop browsers in the compliance office and from handheld tablets and PDAs already in use on the receiving dock — no new dedicated hardware required.
8. Why is the platform deployed on-premises instead of the cloud?
To keep material and supplier data inside the client's own infrastructure, since MTRs and EBOMs often contain commercially sensitive supplier and specification information.
9. How does heat-level traceability help during a supplier quality issue?
Because every component is traced back to its source MTR and heat number, the compliance team can scope exactly which components were affected by a supplier issue within minutes, instead of the days a manual document search would take.
10. How does Atomic Loops build AI systems for compliance-heavy manufacturing environments?
By designing around how compliance teams actually work — multi-language extraction, automatic cross-checks against engineering requirements, and a full audit trail — rather than adapting a generic document-scanning tool to a regulated environment.
Industry Services
Education
Focus areas - Education
© 2026 London School of DIGITAL BUSINESS.
All Rights Reserved