Reading theRight Pages
How a legal team used MiruIQ as contract analysis software to extract specific clause data from 2,000+ contracts without reading a single page.
80,000+ pages reduced to structured findings per clause type. Due diligence timeline compressed dramatically. Every finding traceable to a specific page.
A corporate legal team conducting acquisition due diligence faces a virtual data room with 2,000+ documents, 30–80 pages each. Associates need specific clauses: change-of-control provisions, indemnification caps, termination rights. Reading every page is a six-week task for four people.
Ingestion
The data room contents are synced into the pipeline via S3 or FTP.
Semantic Classification
Each document is evaluated by multiple Semantic Classifiers in parallel, one per clause type the legal team is looking for. Each classifier is configured with a description and examples of what that clause looks like in practice: a Termination classifier knows what termination language looks like across different contract styles, a Change of Control classifier recognizes acquisition triggers and consent provisions, an Indemnification classifier identifies cap language and carve-outs. Only documents that semantically match a clause type proceed; everything else is skipped.
Clause Structure Classification
Each Semantic Classifier feeds its own Structure Classifier, which defines the specific fields to extract for that clause type. The Termination structure defines fields like notice period in days, effective date after notice, and penalty clauses. The Change of Control structure captures trigger conditions, consent requirements, and assignment rights. The Indemnification structure specifies cap amounts, carve-outs, and survival periods. This turns legal document extraction into a schema problem: each clause type is a structure, and contract data extraction fills that structure field by field.
Extraction
Each Structure Classifier feeds its own Extractor, which pulls the actual values from the document pages. The Termination extractor reads the notice period and post-notice duration from the relevant paragraphs. The Change of Control extractor captures trigger events and consent requirements. The Indemnification extractor pulls cap amounts and survival periods. Each extractor outputs structured data ready for downstream systems.
Output
The extractors produce two outputs. The source documents are stored in S3, so associates can click through to the original when they need to verify a finding. At the same time, the extracted values are written to PostgreSQL as a due diligence matrix: for each document, which clauses were found, on which pages, with key terms and risk flags. Associates work from this matrix instead of reading raw documents. When a new clause type becomes relevant (e.g., non-compete provisions), a new Semantic Classifier, Structure Classifier, and Extractor triplet is added without changing anything else: legal document automation that grows with the deal.
Semantic-First Filtering
Multiple semantic classifiers run in parallel, each configured with descriptions and examples of a specific clause type. Only documents that semantically match proceed; everything else is skipped before any structure or extraction is applied.
Parallel Structure Expansion
Each semantic classifier feeds its own structure classifier. The semantic classifier decides if the clause is present; the structure classifier defines what fields to extract from it. New clause types are added as new classifier pairs without changing existing ones.
Paired Classification-Extraction
Each structure classifier feeds its own dedicated extractor. The structure defines what to look for; the extractor pulls the actual values. The pair operates as a unit, reusable across different pipelines.
Semantic-first, per-clause pipelines: each clause type has its own semantic classifier (with examples), structure definition, and extractor. The triplet operates as a unit, and new clause types are added without touching existing ones.
Contract Questions
What legal teams ask about automated contract analysis.
Anything you can define as a schema: parties, dates, notice periods, termination clauses, change-of-control conditions, indemnification caps. Each clause type becomes a structure and legal data extraction fills it field by field, across an entire data room, not one contract at a time.
By keeping the AI where the documents are. MiruIQ runs on the firm's own infrastructure or in Swiss hosting with local models and per-pipeline state isolation; no document, embedding or extraction result ever resides with a third-party AI provider.
See a Pattern That Fits Your Workflow?
Every pattern is built from the same modular components: classifiers, extractors, validators, and connectors. Let's find the right combination for your documents.