Document Data Extraction Specialist

The work

You will read PDF and DOCX files across legal, financial, accounting, and compliance topics, then extract information into an annotation panel using a defined JSON schema. You will work with documents such as contracts, patents, financial records, and other business records.

Each task includes three AI-generated draft outputs. You will choose the strongest draft, check every field against the source document, correct errors or missing information, and complete the final output.

• Read source PDF and DOCX documents carefully before extracting information.
• Follow JSON schemas, including required and optional fields, data types, arrays, objects, and nested structures.
• Select and edit the best AI-generated draft instead of starting from scratch.
• Capture all relevant repeated or changing entries, such as line items and multiple clauses.
• Set genuinely missing required fields to null when the schema requires it.
• Write original summaries in your own words when the schema calls for them.
• Check the completed output for accuracy and completeness.

What it pays and takes

This is an intermediate-level, part-time contractor role. The work is open worldwide and involves legal, finance, accounting, compliance, data analytics, and JSON-based document extraction.

• Pay: $8.00 to $11.20 per hour, with a maximum listed rate of $11.20 per hour.
• Time: Less than 20 hours per week.
• Location: Worldwide.
• Work type: Part-time contractor.
• Experience level: Intermediate.
• Documents: PDF and DOCX files.
• Useful knowledge: Legal contracts, patents, financial records, compliance documents, accounting records, and JSON schemas.

How it works

Apply on OpenTrain. The employer reviews applications there.

About AI training work

AI training work uses carefully prepared examples and human reviews to help artificial intelligence systems produce more accurate results. People with legal, finance, accounting, and other subject knowledge are needed because they can spot errors that general reviewers may miss.