What is Meibel Document Intelligence?

Document Intelligence is the document understanding layer that turns complex files into structured, queryable, scored, and traceable context. It understands layout, preserves tables, classifies images, extracts domain-specific metadata, scores confidence on every output, and makes the result available through semantic search, SQL, and graph traversal.

How is it different from OCR?

OCR reads text. Document Intelligence preserves layout, tables, images, metadata, element relationships, confidence, and provenance so the extracted output can be used reliably downstream. Every element gets the processing path it needs: tables stay structured, charts become data, handwriting gets read, diagrams get described.

How is it different from basic RAG?

Basic RAG chunks and retrieves text. Document Intelligence combines structure-preserving ingest, semantic retrieval, SQL over extracted data, graph traversal across document relationships, and confidence scoring. An agent can use all three retrieval modes in a single reasoning step.

How is it different from cloud document AI (Google Document AI, Azure Document Intelligence, AWS Textract)?

Cloud document AI services parse individual files. They do not build corpus-level connections, consolidate schemas across documents, construct reference graphs, or integrate confidence scoring into execution routing. Meibel does all of that and includes the retrieval layer so you do not need to assemble a separate stack.

Can it handle tables and scanned documents?

Yes. Tables are preserved as structured, queryable data. Image-based tables get OCR plus schema inference. Charts are processed through vision models to extract data points. Diagrams get visual descriptions. Handwriting, stamps, and annotations are handled automatically.

Can teams define their own fields?

Yes. Seven built-in metadata models cover insurance, legal, medical, product/manufacturing, construction, bibliography, and custom domains. Teams can also define custom extraction schemas per data source, industry, workflow, or domain need.

What happens when confidence is low?

Low-confidence outputs can trigger retry, re-extraction, review, escalation, or blocking based on field importance and workflow rules. Reviewers see the original document alongside extracted data with bounding box overlays and field-level status indicators.

How does provenance work?

Every extracted value traces back through the pipeline to the source document, page, and exact region where the evidence came from. Bounding box coordinates identify the precise location. At the corpus level, the citation graph traces relationships between documents: which documents reference which, which components were derived from which sources.

Document Intelligence is free to start. Production usage is billed per page, starting at $0.03/page with volume pricing available. LLM costs are passed through at cost with no markup. Contact us for volume pricing.

How long does setup take?

Minutes for your first document parse. Days for schema configuration and extraction tuning on your specific document types. One to a few weeks for full production deployment depending on output complexity. POC at no charge to ensure product fit.

Can it run in secure environments?

Meibel supports SaaS, bring-your-own-cloud (data plane in your environment), and on-premises deployment. For BYOC, documents never leave your environment. Client-side processing for regulated industries where documents cannot leave the user's device is on the roadmap.

AI builders, platform teams, data teams, product operations, risk and compliance teams, and enterprises building AI workflows on complex document collections.

CONFIDENCE

You don't hope AI is right. You measure it.

A fluent answer and a correct answer look identical. Confidence turns that into a number you can act on: how well the evidence supported the output, how the retrieval behaved, whether the tool call was the right one.

WHY CONFIDENCE MATTERS

Outputs sound equally confident whether they are right or not.

No way to score what an AI output is based on. Teams are stuck between two bad options: automate blindly or review everything manually. These are the symptoms.

Fluency reads as accuracy

The confident tone is identical whether the answer is grounded or invented.

Failures blur together

Retrieval, tool calls and generation all fail into the same bad answer, with nothing separating them.

One score for the whole output

A document-level number tells you something is wrong somewhere, which is the same as telling you to check all of it.

A number with no reasoning

A score you cannot interrogate is a score you cannot act on, and reviewers learn to ignore it.

Offline evaluation, online behaviour

Test-set results say little about what the system does on the long tail it meets in production.

Silent Quality Drift

Embedding changes, data updates, and workflow edits degrade output quality over time.

HOW CONFIDENCE WORKS

Score the Output. Then Act on the Signal.

Confidence is not a model guess. It is a measurable signal derived from execution. Meibel evaluates confidence by measuring evidence quality, context completeness, and step-level behavior, then uses those signals to determine what proceeds, what retries, and what stops.

Confidence Evaluation

Acting on Confidence

Evidence Scoring

Score How Well Each Output Is Supported

Every output receives a confidence score based on the quality and relevance of the evidence behind it. Meibel evaluates whether the retrieved context actually supports the claims in the output, not just whether context was present. This is not a model self-assessment. It is an independent evaluation of how well evidence maps to the generated result.

AI Agents

LLM Hardrails

AI Runtime Orchestration

Agent Control & Runtime Governance

You define the agent: instructions, data sources, tools, and output schema. Hardrails enforce exactly what the agent can see and do at runtime, before the model call, during data access, and at every tool execution.

Agent definitions versioned, immutable
Execution policies enforced before the model runs
Approval gates on any step
One agent, unlimited scopes

AI Confidence Scoring

Source Tracing

Drift Detection

Confidence You Can Measure

Meibel scores every extraction and every agent step across independent dimensions: Correctness, Coherence, Completeness, Faithfulness, Relevance, and OCR Confidence. Scoring is built into execution, not bolted on after.

Evidence-Backed Decisions
Outcome Provenance
Decision-Aware Automation

CONFIDENCE SIGNALS

Every Layer. Every Step. One Consistent Score.

Confidence is evaluated across retrieval, context assembly, tool execution, and model generation. Scores are consistent regardless of which model, tool, or data source is involved. When any component changes, the confidence layer detects the impact. No need to rebuild the evaluation framework.

WHAT ENTERPRISE AI TEAMS BUILD WITH MEIBEL

AI Agents for Industries That Run on Complex Data

Manufacturing and Industrial Distribution: Supplier Document Extraction and Quality Monitoring

Extract product specifications, COA values, batch numbers, and test methods from thousands of supplier documents in different formats. Run trend analysis across extracted data to catch quality shifts over time. One manufacturer processes 30,000 COAs per month and uses agents to build a queryable product information system across 11,000+ technical documents.

Construction and Engineering: Spec Extraction to Inspection Automation

Extract inspection requirements from 1,600+ page specification documents. Match requirements to SOPs by following references from spec to SOP to inspection template. Push structured data to project management systems. One construction company uses agents across 100+ projects per year to automate the chain from specs to inspections to accountability.

Insurance and Financial Services: Claims Document Processing and Validation

Extract policy details, financial fields, and coverage metadata from carrier statements and claims documents. Validate extracted data against structured policy tables. Score confidence on every financial value. One insurance software company reduced document processing from minutes to seconds while scaling from 110 to 30,000 users.

Legal, Compliance, and Government: Regulatory Translation and Compliance Mapping

Translate regulatory documents into compliance policies. Follow the reference graph from a policy clause to the regulation it derives from, and cross-reference regulations with structured compliance data. Agents produce auditable outputs with per-field provenance back to the page and region each value came from.

Enterprise Knowledge Bases: Document Q&A with Source Citations and Confidence Scores

Connect your knowledge bases and let teams ask questions with source citations, confidence scores, and precise numerical answers. Combines search by meaning, structured query for the numbers, and graph traversal across citations, in one reasoning step. Works across legal agreements, technical manuals, formulation documents, and RFP libraries.

Document Corpus Analysis: Schema Discovery Before You Build

Analyze a document corpus and propose extraction schemas before you build. The agent reviews your documents, identifies common fields, groups document types, and produces a discovery report. One customer used this to analyze 200+ files and define a normalized schema before writing a single extraction rule.

CRM and System Integration: Structured Data Extraction to Downstream Action

Process CRM data, call external APIs, route actions based on confidence scores, and push results to downstream systems. This pattern works without Document Intelligence when your data is already structured or accessible through APIs and tools.

Services Partners and SIs: Reusable Agent Patterns Across Multiple Clients

Build reusable agent patterns that deploy across multiple clients. Flexible rules per client. Clear separation of logic and data. One partner identified 152 automatable tasks across five workstreams for a single client and delivers 2-3x faster than building from scratch.

Frequently Asked Questions

Is this just model confidence or log probabilities?

Retrieval by meaning is one of three modes, not the whole system. Standard RAG matches a query against chunks and returns passages. A Meibel datasource also keeps your structured data queryable by its columns and values, and keeps the references between documents traversable, and the agent chooses which mode fits the question. The platform recovers structure during ingestion rather than relying on similarity alone.

Can confidence stop an answer from being returned?

No. The datasource is separate from the agent definition. Agents bind to it, so changing a model or publishing a new agent version does not touch the prepared data.

Does confidence work across different models?

Tables are rebuilt as grids with rows, columns and spanning cells, and stay addressable by cell position. Line and scatter charts are digitised back into series data rather than described in prose.

How does Meibel detect quality drift?

25+ formats including PDFs, office documents, images and scans, email and archives. Medical records are supported. CAD and blueprints are not.

What is the difference between evidence scoring and context completeness scoring?

There is no practical page ceiling. Documents up to 1 GB, and that limit is raisable per customer through the API.

Does this replace my evaluation or testing framework?

No. Meibel orchestrates how data is prepared and used for AI without replacing your storage. It adds structure, metadata and provenance between your sources and your model calls.

How does Confidence relate to Context and Control?