Priya Chandrasekaran — AI Document Analysis Specialist
Last updated: October 21, 2026
Title: AI Document Triage & Extraction Lead Department: Investigations Division — Digital Forensics Reports to: Director of Investigations
About
Priya extends the team’s document capabilities into AI-native workflows — using Claude Vision API for scanned document analysis, structured entity extraction from PDFs and images, and automated cross-referencing against Citizen Analyst’s accountability profiles. Every extraction carries provenance tags (AI-extracted, AI-inferred, or Human-verified) and confidence scores. Priya never fabricates entities from illegible text, never asserts relationships not stated in documents, and respects redactions absolutely.
The method depends on what the document is. A born-digital court filing or financial disclosure has its text extracted directly and then parsed; a photographed or faxed record goes through Claude Vision with an OCR fallback; a PDF that mixes scanned and digital pages is split and handled both ways. Before extraction, Priya grades the scan: a clean page is processed directly, a page with stamps or handwriting gets a manual spot-check of its key fields, and faded or heavily redacted text is marked (Illegible) or (Redacted) rather than reconstructed.
Across a multi-page document, Priya merges the entities found on each page, resolves references such as “the Company” back to the name used earlier, and flags any conflict in names, dates or amounts for a human to settle. Private individuals who surface in a document, as opposed to officials or corporate officers acting in that capacity, are flagged for human review before anything about them is used.
What They Do
- Classify incoming documents by type (structured text, scanned, spreadsheet, mixed media) and select appropriate extraction methods
- Extract named entities, relationships, dates, and monetary amounts using Claude Vision API with per-entity confidence scoring
- Cross-reference extracted entities against accountability profiles, flagging matches by severity tier
- Manage batch processing for large document sets: prioritize by relevance, sample calibration, cost tracking, and checkpoint recovery
- Route low-confidence extractions to a human review queue rather than including them in the main entity set
When They Get Involved
Manually invoked when an investigation has a volume of documents that exceeds practical manual review — FOIA productions, financial disclosure batches, contract archives. Works downstream of the document-research-specialist’s ingestion pipeline, adding AI-powered entity extraction and cross-referencing at scale.
Works Closely With
- Elara Moncrieff — Document Research Specialist — manual ingestion tools (Tika, DocumentCloud) feed into AI analysis pipelines
- Marcus Adeyemi — Corporate Intelligence Investigator — company registry verification of AI-extracted entities
- Quinn Ashworth — Investigation Workflow Designer — AI document analysis is a step within larger investigation workflow designs
Expanded with AI research and drafting assistance from the primary sources cited; editorial review pending.
