Use cases are assembled from pipelines that already exist.
A builder does not start from a blank canvas. The steps below are published, versioned and tested, and a new use case is mostly a matter of wiring them in the right order and adding the step that is specific to the job.
Retrieval
OCR
Scanned pages and images turned into text with the page kept.
Web search
Public sources searched and fetched with the URL and date kept.
Filing retrieval
Filings pulled by company and period, with the page kept for each figure.
Licensed data retrieval
Licensed sources read with the license terms tracked on each result.
Detection and analysis
Classification
Documents or records sorted into the types the pipeline expects.
Financial calculators
Multiples, ratios and covenant tests computed as code.
Documents
Document parsing
Sections, tables and fields pulled out of a document with their location kept.
Review
Review agent
A checklist applied to a draft or a record, with each item passed, failed or sent to a person.
Evidence scoring
Each claim scored on source quality and agreement, with the citation kept.
Entitlement checks
Each read checked against the user’s entitlements and the information barrier.
Evaluation sets
A test set with known answers kept with the pipeline and rerun on each version.
Output
Templates
A company template filled from pipeline output, with each field traced.
HTML reports
A report or brief built from the run, with citations and the run record.