Promeing ConceptScenario set in Tirana
Business Operations Platform
IndustryDistribution and trade
Visualisation
- Status
- Promeing Concept
- Scenario
- Tirana
- Industry
- Distribution and trade
- Capabilities
- AI AutomationAI AgentsAI Products
- Sheet
- 03 / 06
Bilingual document intelligence for a distributor: invoices, orders, contracts and customs documents become structured records, reconciled and routed, with an assistant that answers from the company’s own records.
A Promeing Concept: a studio study of a system we would build, set in a realistic scenario. It describes no client, no deployment and no measured result.
Problem
A growing distributor works across Albanian and English, on paper, spreadsheets and messaging apps. Invoices are typed in by hand, orders are confirmed in chat threads, and the answer to “where is this order?” depends on who you ask.
Context
The scenario: a trading and distribution company in Tirana, buying from suppliers abroad and selling to retailers across the region. Documents arrive in both languages, as scans, photos and PDFs, and customs paperwork adds a third format of its own. Off-the-shelf tools read English well and Albanian poorly, so the work falls back to people re-keying what is already written down.
System
Document intelligence and workflow automation read invoices, orders, contracts and customs declarations into structured records. Each extracted field keeps a link to where it was found on the page. Records are reconciled against stock receipts and payments, mismatches are routed to the right person for approval, and staff get an assistant that answers questions such as “where is this order?” from the company’s own records, with sources. It is designed bilingual from the first line of code; the technical interest is reliable extraction for Albanian, a lower-resource language, which a pilot would evaluate on the company’s real document types, used with consent.
AI architecture
Documents enter through one intake, are classified and extracted into records, and are reconciled against the ERP. An approval workflow handles exceptions, and the assistant answers only from records it can cite.
AI architecture of Business Operations Platform
Layers
- Interface: Document intake, Staff assistant
- Agents: Approval workflow, Reconciliation agent
- Retrieval: Record search with sources
- Models: OCR and layout model, Bilingual extraction model
- Data: Structured records
- Infrastructure: ERP and accounting integration
Connections
- Document intake → OCR and layout model
- OCR and layout model → Bilingual extraction model
- Bilingual extraction model → Structured records
- Structured records → Reconciliation agent
- ERP and accounting integration → Reconciliation agent
- Reconciliation agent → Approval workflow
- Structured records → Record search with sources
- Record search with sources → Staff assistant
Interface
The operations view: a bilingual document stream on the left, the structured record in the middle with every field traced to its source, and an assistant that answers in either language, with sources.
- Faturë · Invoice DOC-0087SQ Extracted
- Porosi · Purchase order DOC-0088EN Matched
- Deklaratë doganore · Customs DOC-0089SQ Needs review
- Kontratë · Contract DOC-0090SQ · EN Queued
- Furnitori · Supplier
- Furnitor Shembull sh.p.k.
- Nr. i faturës · Invoice no.
- FT-2026-00412
- Data · Date
- 2026-09-28
- Porosia · Order
- PO-2026-0418
- Përshkrimi · Description
- Çokollatë e zezë, kuti
- Shuma · Amount
- —
- Order PO-2026-0418Matched
- Stock receipt GR-2026-0331Matched
- PaymentNot yet due
Route: approval by finance
Porosia u pranua në magazinë më 2026-09-30. Fatura FT-2026-00412 përputhet me porosinë.
Received at the warehouse on 2026-09-30. Invoice FT-2026-00412 matches the order.
- A: Documents in Albanian and English arrive in one stream and are classified on arrival.
- B: Every field is extracted with a link to where it was found, then reconciled against orders and stock.
- C: The assistant answers in the language it is asked, and only from records it can cite.
Numerals appear only in dates, IDs and timestamps. Indicators read — until there is something true to show.
Technology
- OCR and layout analysis for scans, photos and PDFs
- Extraction models evaluated per document type and per language
- Albanian and English language support, including ë and ç in every field
- Workflow engine for approvals and exceptions
- Reconciliation rules against stock receipts and payments
- Hybrid vector + keyword search over structured records
- ERP and accounting integration through existing APIs or exports
Expected outcome
ExpectedLess manual re-entry, fewer orders lost between chat threads, and one searchable record of how the business runs, in both of its languages.
Qualitative and unmeasured. A pilot measures it against acceptance criteria agreed at the start.
What we would measure
- Extraction accuracy per document type and language
- Manual entry hours removed
- Exceptions resolved within the approval workflow
- Citation accuracy of the assistant
- Time to answer “where is this order?”
Metric names only. Values exist once a pilot measures them.
What a pilot would need
- A consented sample of each document type, in both languages
- Read access to the ERP or accounting exports
- The reconciliation rules the finance team already applies
- One approver per exception type
- A shared inbox or scanner as the single intake
- Agreement on where the data is processed and stored
Promeing ConceptScenario set in Tirana
Building something like this?
Bring the problem in its messiest form. We’ll tell you honestly whether AI is the right way to solve it.