Freelance › Projects › Software development › OCR pipeline to turn scanned invoices into structured data
OCR pipeline to turn scanned invoices into structured data
Employer
Andriy
Project parameters
Type of cooperationOne-time project
SectionSoftware development
Prepaymentwithout prepayment
Payment methodsCash, Bank transfer
Acceptance of requestsfrom today, 20:28 until Aug 30, 2026
Project description
Our company has accumulated tens of thousands of paper invoices and scanned documents that currently sit in folders and PDF archives with no way to search or analyze them. We need a working OCR document-digitization pipeline that takes a batch of scans, recognizes the text on each page, and turns unstructured images into clean, structured records we can query. The volume grows every week, so the solution must run unattended on a schedule, pick up new files from an input folder, and process them reliably without a person babysitting the queue. Accuracy matters more than raw speed: a wrong total or a misread date creates real accounting problems downstream, so the pipeline has to be conservative and flag anything it is unsure about.
Concretely, for each document we need the system to extract the key fields we care about: supplier name, invoice number, issue date, line items where present, currency, subtotal, tax and grand total. After extraction the values must be validated, dates parsed into a single canonical format, numbers checked so that line items reconcile against the stated total, and low-confidence results routed to a review queue instead of being written silently. Clean records should be exported both to our database and to a spreadsheet so the finance team can work with them immediately. We would prefer a Python implementation with a well-known OCR engine, clear logging, and a simple way to re-run failed documents.
Please treat this as a production tool, not a proof of concept. We will provide a representative sample set of real scans (with varying quality, rotation and languages) for you to tune against, and we expect the deliverable to include the source code, setup instructions, and a short handover so our own staff can operate and extend it.
— Ingest scans and PDFs from a watched folder, deskew and clean images before recognition
— Extract supplier, invoice number, dates, totals, tax and line items with confidence scores
— Validate and normalize fields; reconcile line items against the total; route unsure results to review
— Export structured records to the database and to a spreadsheet; provide logs and a re-run option
Concretely, for each document we need the system to extract the key fields we care about: supplier name, invoice number, issue date, line items where present, currency, subtotal, tax and grand total. After extraction the values must be validated, dates parsed into a single canonical format, numbers checked so that line items reconcile against the stated total, and low-confidence results routed to a review queue instead of being written silently. Clean records should be exported both to our database and to a spreadsheet so the finance team can work with them immediately. We would prefer a Python implementation with a well-known OCR engine, clear logging, and a simple way to re-run failed documents.
Please treat this as a production tool, not a proof of concept. We will provide a representative sample set of real scans (with varying quality, rotation and languages) for you to tune against, and we expect the deliverable to include the source code, setup instructions, and a short handover so our own staff can operate and extend it.
— Ingest scans and PDFs from a watched folder, deskew and clean images before recognition
— Extract supplier, invoice number, dates, totals, tax and line items with confidence scores
— Validate and normalize fields; reconcile line items against the total; route unsure results to review
— Export structured records to the database and to a spreadsheet; provide logs and a re-run option
