Every employer and software package prints a payslip differently. Fields sit in different places under different labels, so one fixed template never covers them all.
Case study · AI payroll tax & management system
AI that evaluates payroll tax for an Italian government body.
AI payroll tax calculation software for an Italian government body: the body could not evaluate taxes from its users' payroll, so it called Eziline. We built SolidaNetwork, an AI-based tax-from-payroll and management system. Citizens upload their documents on IRIS, the system reads them and evaluates the tax automatically, and the body now manages its taxes effectively.
8Document families the engine is configured to read
12Layout variants recognised automatically across four of them
100+Fields mapped on a single document type
- Client
- An Italian government body (name withheld)
- Industry
- Public-sector tax administration, Italy
- Problem
- Tax could not be evaluated from citizens' payroll documents
- Solution
- SolidaNetwork: AI document reading, tax evaluation and management
- Citizen access
- Documents uploaded on IRIS
- Delivered by
- Eziline Software House, Islamabad, Pakistan
The challengePayroll arrives in many layouts Data trapped in PDFs Evaluation depended on structure
The numbers were there. They were locked inside documents.
To evaluate a person's tax from payroll you need the figures on their payslips, certificates and returns. For the body, those figures arrived as documents in many shapes, and nobody could turn them into a tax evaluation at scale.
The amounts that matter are printed in documents. Retyping them is slow, expensive and the easiest place to introduce an error that later becomes a wrong tax figure.
A tax rule can only run on clean, labelled data. Without a reliable way to read the documents, the body could not evaluate taxes from payroll and could not manage them with confidence.
What the system reads
One engine, eight document families.
SolidaNetwork's extraction engine is configured for the Italian documents that carry payroll and tax information, plus the supporting financial papers that often travel with them. In the engine they are configured under their Italian names: busta paga, Certificazione Unica, Redditi, cartella, estratto conto, CRIF, centrale rischi and visura protesti.
Payroll and income-tax documents
Payslips · Certificazione Unica · Income tax return · Tax collection notice
Supporting financial documents
Bank statements · Credit report · Credit-register extract · Protest search
What was builtCitizen upload on IRIS Document-type handling Layout recognition Layout-aware reading Rule-driven field extraction Table extraction AI tax evaluation from payroll Spreadsheet-ready output Guardrails on every upload Management for the body Service architecture
From a citizen's upload to an evaluated tax.
SolidaNetwork joins a citizen-facing upload, a document-reading engine and the tax evaluation and management layer, so the body works from structured data instead of paper and PDFs.
Citizens are given access to upload all their documents on IRIS. That single entry point replaces collecting papers by hand, and everything that follows starts from what the citizen submitted.
IRISCitizen accessOne entry point
The engine is configured for eight Italian document families: payslips, Certificazione Unica, income tax returns, tax collection notices, bank statements, credit reports, credit-register extracts and protest searches. A new family is added as a new configuration.
8 document familiesItalian documentsConfigurable
Payslips, tax returns, Certificazione Unica and protest searches each come in several printed layouts. The engine looks at the top lines of the first pages and matches the document to the right layout variant, 12 variants in all.
12 layout variantsTemplate matchingPayslip · CU · Redditi
Every word is read together with its position on the page. Words are grouped into lines and, for payslips, into blocks, so a value is understood by where it sits and what label is beside it, not just by the raw text.
Word coordinatesLines and blocksPDF text layer
Each document family has its own rule set: label-and-pattern matching, values found between two printed markers, values above or below a label, and combinations of fields. The rules live in configuration, so a layout change is a rule edit rather than a rewrite.
Pattern rulesBetween-label rulesConfig-driven
Bank statements, tax collection notices and credit-register extracts carry long tables. The engine rebuilds the rows and columns, normalises amount formatting and drops duplicate rows before output.
Rows and columnsAmount clean-upDe-duplication
Once the figures are structured, the system evaluates the tax automatically. Applicable payroll and income-tax rules are applied to the extracted data, so the body gets an evaluation rather than a pile of documents.
Payroll to taxRules appliedAutomatic
Each document becomes an Excel workbook: named fields in order, tables on their own sheet where a document has one, and amounts in a form a spreadsheet can read. Staff can open, check and reuse the result at once.
Excel outputNamed fieldsTable sheets
Only PDF files are accepted. A document that does not match a known layout is reported as a new template, not guessed at. A document type that has not been configured is refused with a clear message. Fields not found are left blank, never invented.
PDF onlyNew-template alertNo guessed values
Evaluations sit in one managed system, so the body can follow what has been submitted and what tax has been evaluated, and manage its taxes from a single place instead of from scattered files.
One systemTax managementOversight
The reading engine runs as a Python web service built on Flask and the PyMuPDF library, with pandas and openpyxl writing the workbooks. It is a separate service, so it can be called from a portal such as IRIS.
PythonFlaskPyMuPDF · pandas
How a document becomes a tax evaluation
Upload, read, extract, evaluate, manage.
Follow one citizen document through SolidaNetwork. Each step feeds the next, and nothing is retyped between them.
- Upload
The citizen uploads their documents on IRIS.
- Recognise
The engine identifies the document family and matches its layout variant.
- Read
Words are read with their positions and grouped into lines, blocks and tables.
- Extract
Rules pull each named field and table into structured data.
- Evaluate
Applicable payroll and income-tax rules are run on the data to evaluate the tax.
- Manage
The body reviews and manages the result in one system.
The pipeline
Document to tax evaluation, in four stages.
Watch a document move through SolidaNetwork: uploaded on IRIS, read and matched to its layout, turned into structured fields, then evaluated for tax and passed to review.
What changedCitizens upload once Tax evaluated automatically Structured data, not scans Taxes managed effectively
A government body that can finally evaluate and manage tax from payroll.
All documents go in through IRIS. Citizens are given access to submit everything in one place.
The system reads the documents and evaluates the tax, so the body no longer depends on retyping figures by hand.
Every document becomes named fields and tables in a spreadsheet-ready workbook that staff can check and reuse.
The body now manages its taxes effectively, from one system built for the job.
In plain words
AI payroll tax calculation software and automated tax evaluation for government
Tax authorities and public bodies sit on a data problem before they have a tax problem. Citizens' income is described in payslips, certificates and returns, each printed in its own layout. AI payroll tax calculation software closes that gap: it reads the documents, turns them into structured data, and runs the applicable payroll and income-tax rules on that data. The result is an automated tax evaluation system that works from what the citizen actually submitted.
SolidaNetwork shows what that looks like for an Italian government body. Citizens use a document upload portal, IRIS, to submit their papers. SolidaNetwork identifies each document, matches its layout, extracts the fields and tables with AI document processing, and evaluates the tax. The same pattern applies to any e-government platform that has to turn citizen documents into decisions, from payroll tax automation to income verification and benefit eligibility.
Eziline is a tax software development company and AI automation company in Islamabad, Pakistan, building for clients abroad as well as at home. We design the document reading, the rules and the management system around your documents and your law, host or hand over the platform, and support it after go-live. If your organisation is buried in forms, we can scope a version for you.
AI payroll tax calculation softwareAutomated tax evaluation systemDocument upload portal for citizensAI document processingPayslip data extractionGovernment tax management softwarePayroll tax automation ItalyE-government platform developmentTax software development companyPDF to Excel data extractionIntelligent document processingAI automation company Pakistan
Questions
SolidaNetwork: AI payroll tax system: frequently asked questions.
What is AI payroll tax calculation software?
It is software that reads payroll and income documents, turns them into structured data and applies the applicable tax rules to evaluate the tax automatically. In SolidaNetwork, citizens upload their documents on IRIS and the system does the reading and the evaluation.
What problem did the Italian government body have?
The body could not evaluate taxes from its users' payroll. The figures were locked inside documents with many layouts. Eziline built SolidaNetwork, an AI-based tax-from-payroll and management system, so the body can evaluate and manage its taxes.
What is IRIS in this system?
IRIS is where citizens are given access to upload all their documents. SolidaNetwork takes those uploads, reads them and evaluates the tax. Eziline can build a similar document upload portal for citizens or customers.
Which documents can the system read?
The engine is configured for eight Italian document families: payslips, Certificazione Unica, income tax returns, tax collection notices, bank statements, credit reports, credit-register extracts and protest searches. New families are added as new configurations.
Does it use OCR?
The reading engine works on the text layer of digital PDFs and uses each word's position on the page. Scanned paper needs an OCR step in front of it, which Eziline adds when a project calls for it.
What happens if a document does not match a known layout?
It is not guessed at. The engine reports a new template so that a rule set can be added, and fields it cannot find are left blank rather than invented. Only PDF files are accepted.
Can this be built for a government or tax body in Pakistan?
Yes. Eziline is based in Islamabad and builds document processing, tax evaluation and management systems around your documents and your rules. Book a free consultation and we will scope it.
Your turn
Need a system like this for your organisation?
Tell us how your teams work today. We will show you the closest running system, then scope and quote yours with your workflows, roles and reports.