Modular Processing,Built for Real-WorldDocuments
MiruIQ is built around modular processing units, called modules. Each module performs a clearly defined task (classification, AI data extraction, or verification) and can be combined freely into pipelines.
Classifier Modules
Classifier modules are MiruIQ's document classification software: they determine what a document is or what it contains, and assign documents to one or more groups when defined conditions are satisfied.
Classification results are used to:
Pipeline Control
control pipeline branching
Extraction Logic
select extraction logic
Routing
route documents to different outputs
Multiple classifiers can run sequentially or in parallel.
Simple Structure Classifier
Identifies documents by matching them against a single, well-defined structure.
The classifier evaluates whether a document contains the required fields, sections, or layout characteristics that define a specific document type, independent of formatting details such as positioning or visual design.
Simple structure classifiers are typically used as the first decision point in a pipeline, establishing a reliable baseline for downstream processing.
They enable consistent handling of documents that share the same logical structure, even when appearance varies over time.
Multi-Structure Classifier
Identifies documents that contain multiple independent structures within a single file.
Instead of treating a document as one flat entity, the multi-structure classifier detects and validates several distinct structures, such as multiple tables, repeated sections, or logically separate data blocks.
Each detected structure can be processed independently, enabling more granular extraction and validation without splitting the document manually.
Multi-structure classifiers are especially useful for complex documents where meaningful information is distributed across different sections or repeated patterns.
Semantic Classifier
Identifies documents based on meaning, rather than structure, layout, or exact wording.
The classifier evaluates the content of a document to determine what it describes, allowing MiruIQ to distinguish between documents that may share a similar structure but differ in intent, subject, or context.
They focus on what a document is about, not how it's presented, remaining effective even when terminology varies or documents evolve.
Semantic classification enables downstream processing, routing, and extraction to be driven by meaning instead of rigid patterns.
File Name Classifier
Identifies documents by matching their filenames against configurable patterns.
The classifier reads the filename of each incoming document and evaluates it against a set of user-defined patterns or naming conventions. When a match is found, the corresponding structure is assigned, enabling the pipeline to route and extract without opening the file.
File name classifiers are ideal when document providers follow consistent naming conventions, making classification fast and deterministic.
They enable instant routing decisions based on metadata alone, keeping processing lightweight and allowing downstream modules to start extraction immediately with the correct schema.
Extractor Modules
Extractor modules are the data extraction tool at the core of MiruIQ, transforming classified documents into structured, normalized data.
An extractor inherits the structure identified by preceding classifier modules and uses it as the foundation for document data extraction. This means extraction always operates on a known, validated structure, not on raw document layout.
OCR data extraction, document parsing, layout analysis, and semantic mapping happen inside the module. By separating structure detection from extraction logic, MiruIQ ensures that data shaping remains stable even as documents evolve.
Field Selection
The extractor allows precise control over which fields are included or excluded.
Users can focus extraction on the data that is actually required, avoiding unnecessary or sensitive information from entering downstream systems.
This keeps data models clean and prevents accidental coupling between document content and application logic.
Field Mapping
Extracted fields can be mapped into a target structure defined by the user.
This includes renaming fields, aligning them to existing schemas, and reshaping data so it fits seamlessly into databases, APIs, or internal models.
By handling this mapping within the extraction step, MiruIQ removes the need for additional transformation logic later in the pipeline.
Nested Data
Extractor modules support nested and hierarchical data structures, including arrays and repeated elements.
Complex document content such as tables, line items, or grouped sections can be mapped directly into structured, nested outputs, preserving relationships instead of flattening data. This is what makes table extraction from PDF reliable: rows, columns, and line items arrive as structured arrays instead of flattened text.
This makes it possible to model real-world information accurately, even when documents represent it inconsistently.
Consistent Output
The extractor produces predictable, consistent output structures across varying document layouts.
Because extraction is based on inherited structure and defined meaning, downstream systems always receive data in the same shape, regardless of how the original document was formatted or ordered.
This consistency is critical for reliable automation and long-term maintainability.
Verification Modules
Verification modules ensure that extracted data is valid, consistent, and trustworthy before it is released to downstream systems.
They introduce explicit control points into automated pipelines, enabling both fully automated and human- or system-in-the-loop decisioning, across individual documents or entire document groups.
Verification can be applied at different stages of a pipeline and combined freely with classifiers, extractors, and connectors.
Simple Value Checker
Validates extracted values against predefined conditions, including semantic equivalence.
This module operates on values that have already been extracted from a document and ensures they meet defined logical or business constraints before further processing continues.
In addition to exact comparisons, the Simple Value Checker supports semantic normalization, allowing different real-world representations of the same value to be treated as equivalent.
Single extracted fields within an individual document.
When validating an address, values such as "Hausstrasse 24" and "Hausstr. 24" are considered equivalent during verification.
API Value Checker
Introduces an external, data-driven verification step into the pipeline.
This module pauses pipeline execution and waits for an external system to provide reference values or thresholds via an API call. Extracted values are then validated against those externally supplied inputs.
This allows verification logic to be dynamic and context-aware instead of hardcoded into the pipeline.
Single extracted fields using externally defined rules or thresholds.
In a loan application process, minimum salary requirements vary per applicant. An external system calculates the required threshold and sends it via API. MiruIQ verifies the extracted salary before continuing.
Simple Cross-Document Checker
Verifies consistency across a predefined group of related documents.
This module operates on document groups with a known, fixed composition. It waits until all expected documents for a group have arrived before performing cross-document checks.
Document groups can be defined in multiple ways, allowing flexible grouping strategies without changing pipeline logic.
Consistency and coherence of extracted fields across a fixed document set.
A loan application must contain an ID document, salary statement, and application form. Once all documents arrive, MiruIQ verifies that applicant name and date of birth match across all files.
API-Based Cross-Document Checker
Performs cross-document verification where document group completion is controlled externally.
Unlike the simple cross-document checker, this module does not rely on a predefined document count. Instead, an external system explicitly signals when a document group is complete.
This enables MiruIQ to integrate with dynamic workflows where document arrival is unpredictable.
Consistency and coherence of extracted fields across a dynamically completed document group.
In a loan management system, documents arrive over time. Once the system determines the application is complete, it signals MiruIQ via API. MiruIQ then verifies identity, address, and financial data before allowing the application to proceed.
Simple Acceptance Gate
Pauses pipeline execution until a user manually confirms or rejects a document within MiruIQ.
This module introduces a human-in-the-loop checkpoint into an automated pipeline. When a document reaches this step, processing halts and the document is presented for manual review inside MiruIQ.
The reviewer can inspect extracted data, verify against the source document, and either approve or reject, resuming or terminating the pipeline accordingly.
Human judgment on document correctness, completeness, or compliance before further processing.
A financial institution processes account opening documents automatically but requires a compliance officer to manually approve each application within MiruIQ before the account is created in the core banking system.
API Acceptance Gate
Pauses pipeline execution until an external system sends an approve or deny signal via API.
This module holds a document at a defined pipeline step and waits for an external decision. A third-party system, review platform, or automation can evaluate the document independently and then call back to MiruIQ with an approval or rejection.
This enables integration with external review workflows, rule engines, or AI-based verification systems without requiring them to be part of the MiruIQ pipeline itself.
External decision on whether a document should proceed, based on logic defined outside of MiruIQ.
A health insurer receives prior authorization requests. MiruIQ extracts the clinical data and sends it to a clinical review system. The review system (staffed by nurses, powered by rules engines, or assisted by AI) evaluates the request and calls back to MiruIQ with an approval or denial, releasing the document to the appropriate workflow.
Designed to Work Together
Automation That Follows Meaning, Not Rigid Rules
MiruIQ automation is built around modular decision steps, not monolithic workflows.
Each module focuses on a single responsibility: understanding structure, interpreting meaning, extracting data, or verifying consistency. Automation emerges from how these modules are combined, not from hard-coded flows.
This makes automation:
Flexible
Adapts to new requirements without rebuilding workflows
Explainable
Every decision step is visible and auditable
Resilient
Changes in one module don't break the pipeline
How Automation Works in Practice
Documents enter the pipeline as raw input: unsorted, heterogeneous, and unpredictable.
From there, MiruIQ applies progressive automation:
Each step reduces ambiguity and increases confidence before the next decision is made.
Branching Instead of One-Size-Fits-All
Automation in MiruIQ is branching by design.
Once a document is classified, it can automatically follow different paths:
This allows a single pipeline to handle many scenarios, without duplicating logic or maintaining separate workflows.
Independent Outputs, One Controlled Flow
Automation does not end in a single result.
In a single pipeline:
Each output is connected to a specific decision step, keeping automation transparent and auditable.
Automation Without Lock-In
MiruIQ does not force automation into a fixed sequence. Modules can be:
This allows automation to evolve alongside business requirements, without breaking existing integrations.
Extraction Questions
What teams ask before evaluating document data extraction software, answered in plain language.
Most tools do one of the three. MiruIQ covers the full sequence in one platform: classifier modules identify the document type, extractor modules pull fields and tables into a JSON schema you define, and verifier modules validate the result, routing low-confidence records to human review instead of passing them downstream.
Schema-driven LLM extraction outperforms template-based OCR on real-world tables: truncated values, multi-page tables, footnotes that redefine what a column means. MiruIQ was built around this case: define the table structure once and the extractor fills it for every document, in a pipeline that scales with volume.
MiruIQ validates on three levels: against the JSON schema, against expected values for a single document, and across documents, comparing names, employers, dates and amounts within a bundle to surface inconsistencies and potential fraud. Records that fail a check park for human review instead of flowing into your systems.
Ready to Transform Your Documents?
See how MiruIQ modules can streamline your document processing workflows.