The work

Each task starts with you sourcing a publicly accessible Bengali PDF in an assigned document type — a newspaper page, a textbook spread, an exam paper, a flyer, form, manual, menu or notice — that contains at least one multimodal element such as a table, diagram, image or handwriting. You then produce a complete structural map of that page: bound every meaningful region, assign it a component type (document title, section heading, paragraph, list, table, figure, caption, formula, question, answer field), give it a reading-order index, and link captions and cells to their parent figure or table. Alongside the structure you transcribe every text region faithfully in Bengali script — including handwritten content — flagging any region where the source is genuinely illegible rather than guessing. Page metadata (language, document type, source, dimensions, flags for tables, formulas and handwriting) is captured with each submission.

Nothing here is model-generated and then cleaned up. Component identification, typing, reading order and transcription are all human-authored, which is the point: the corpus targets exactly the material parsing and vision-language models handle worst in Indic scripts.

What the screen looks for

  • Native Bengali with full script command — diacritics, conjunct forms, numeral variants. A single wrong diacritic counts as a defect, so the screen probes orthographic precision rather than conversational fluency.
  • Document-handling background — annotation, transcription, translation, localization, subtitling, proofreading, journalism, typesetting, or regional-language data review.
  • Taxonomy discipline — whether you can apply a fixed component vocabulary consistently across hundreds of pages instead of improvising a new judgment per document.
  • Comfort with hard layouts — multi-column newspaper pages, exam papers with answer fields, handwritten forms, mixed-script content.

Logistics

Fully remote and asynchronous, open only to contributors based in India, the United States, Canada or Western Europe. Work is task-based with no fixed shifts; accuracy is the first metric, with handling time tracked alongside it. The observed rate for this listing is $13/hr — as posted by the platform, not a guarantee, and pay on Mercor projects varies with task type, review role and assessed quality. Expect a paid or unpaid qualification task involving a real Bengali page before regular volume begins, and expect your early submissions to be reviewed closely by a second Bengali expert.