Configurable document intelligence

Parse any document into trusted data.

Upload documents, classify the right category, extract fields against JSON Schema, validate outputs, and keep humans in control before data is approved.

Workspace / Kinshasa pilot

48 documents processed this week.

Review ready

Pending review

Approved

Schema versions

liste-eleves-goma-2026.pdf

student_list / Validation

Needs review

attendance-week-18.jpg

attendance / Extraction

Running

supplier-registration.pdf

configurable_form / Review

Ready

Staged pipeline

Each step is visible, validated, and recoverable.

The platform does not ask one giant prompt to decide everything. It uses a staged workflow with traceable outputs and review-ready state.

Upload

Store the original file and create a durable document record.

OCR

Extract text while preserving the raw OCR response for audit.

Classify

Match the document to a configured category and schema version.

Extract

Generate structured output against the category JSON Schema.

Validate

Flag schema errors, missing fields, confidence, and warnings.

Review

Let humans edit, approve, reject, or send documents back.

Admin AI playground

Test prompts and schemas before they touch production files.

Compare models, inspect OCR text, validate JSON Schema output, and tune category behavior without hardcoding provider-specific logic.

Schema test result

Category: configurable_form / Version: 3

JSON Schema validation

Passed

Unknown fields preserved

extra_fields

Provider metadata captured

Tracked

Category-driven

Start with school records. Expand to any structured document.

Education is the first market, but categories are configuration: prompts, JSON Schema, validators, field mappings, and review forms can evolve per workspace.

School records: Student lists, teacher lists, attendance sheets, and report data.
Operational forms: Any internal form with known fields, mappings, and validators.
Invoices: Line items, supplier details, totals, dates, and extra fields.
Contracts: Extract parties, terms, dates, obligations, and review warnings.

Build the review layer before automating decisions.

Keep raw OCR, raw AI output, normalized data, validation results, and human corrections together for a safer path to automation.