Image Text Extractor
Extracts structured data from screenshots, images, and scanned documents into row-based records.
What it is
Valuable prospect data does not always arrive in structured formats. Conference lists come as PDFs. Org charts are shared as screenshots. Business cards sit in a rep's camera roll. This agent converts all of it into clean, structured data rows using vision AI.
How it works
The agent accepts image inputs including PNG, JPEG, PDF page renders, and screenshots. It identifies the content type: table, list, form, business card, or document, and applies the appropriate extraction schema. For tables, it reconstructs column headers and row values. For business cards, it extracts name, title, company, email, and phone. Output is written as structured rows to a spreadsheet or JSON, with a confidence score per extracted field and flags for low-confidence extractions requiring human review.
How it fits into the pipeline
Feeds the Ops layer's normalization and quality pipeline. Extracted records flow into the Data Normalisation Agent and Deduplication Agent before entering the main pipeline as account or contact records. Particularly valuable in field sales and event-driven prospecting contexts.