AI Scientific Figure Data Extraction Engineer

Scirvex Intelligence

Remote · $100k – $130k / year · Contract · Posted 4 weeks ago

This role is no longer accepting applications

Don't miss the next role like this

Create a free profile with your skills. We match you to new AI roles as they open, and companies hiring can contact you directly.

Create free profile, 2 minutes

Similar open roles you can apply to now:

See all open AI jobs

This role is no longer accepting applications

Don't miss the next role like this

Create a free profile with your skills. We match you to new AI roles as they open, and companies hiring can contact you directly.

Create free profile, 2 minutes

Similar open roles you can apply to now:

See all open AI jobs

About the role

Most of the quantitative evidence in a research paper lives inside its charts, not its text. Researchers using our systematic review platform keep asking us to pull exact values from bar charts, Kaplan-Meier curves and forest plots so they can run meta-analyses without manual digitizing. Text extraction we already handle well; figures are the gap. On this six-month contract you'll build a pipeline that detects plot types, reads axes and legends, and converts marks into structured tables with confidence scores. You'll work alongside two NLP engineers and a methodologist who will tell you, bluntly, when an extracted number would mislead a review. Extension or conversion to a permanent role is likely if the pipeline lands well.

Fully remote.

Responsibilities

Build a chart-type classifier and data-point extraction models for common scientific plots
Combine OCR on axes and legends with visual models to recover numeric values
Produce confidence scores that tell reviewers which values need a human check
Benchmark accuracy against a hand-digitized gold set from published reviews

Requirements

Hands-on experience with document AI or chart understanding models
Strong Python and PyTorch, plus familiarity with Hugging Face vision-language models
Comfortable with messy PDFs and inconsistent figure styles
Available for at least 30 hours a week overlapping UK afternoons