PDF Annotation & Transcription Experts: Telugu

What you'll do

Document AI still stumbles on Telugu: its conjunct-heavy script, newspapers set in tight columns, and handwritten forms are exactly what parsing models get wrong. This project builds a human-made training set to fix that, one PDF page at a time, and Telugu is one of five Indic scripts in it alongside Japanese and Korean.

Pages are supplied to you; you do not hunt for documents. For each one you:

1. Screen it: confirm it is Telugu, legible, has real content and shows no personal details. Flag unsuitable pages straight away rather than half an hour in
2. Map the structure: draw a bounding region around every meaningful element (title, heading, paragraph, list, table, figure, diagram, caption, formula, question, answer field), give it a type, and number it in reading order
3. Link parts: attach captions and sub-elements to the table or figure they belong to
4. Transcribe: reproduce every region character for character in Telugu script, not transliteration, handwriting included, and flag anything illegible
5. Record metadata: language, document type, source, page size, and whether tables, formulas or handwriting are present

Source material is deliberately mixed: newspapers, textbooks, exam papers, and everyday items like flyers, forms, manuals, menus and notices. No step is machine-generated; every region, type and transcription is done by a person, and each task is then reviewed end to end by a second Telugu expert. Experienced annotators move into that reviewer seat.

Who fits

• Native Telugu speaker with full command of the script, its diacritics and conjunct forms
• A background working with documents: annotation, transcription, translation, localization, subtitling, proofreading, journalism, or regional-language data review
• Precise at the character level: here one wrong vowel sign counts as a mistake
• Consistent: the same taxonomy applied the same way across hundreds of pages

Helpful extras: prior AI data work in Telugu, MTPE or OCR post-editing, typesetting or digitization, and a working knowledge of Unicode normalization and Telugu input methods.

What it pays

$12.68 per hour. That is identical across the Indic versions of this project (Gujarati, Bengali, Malayalam, Odiya), and roughly a third of the $33.58 rate on the Korean version, which is open to different countries. The ad notes that accuracy is the first quality measure, with handling time also tracked, so the effective rate depends partly on how fast you can work without errors.

Payments are weekly on Stripe or Wise. Mercor states it cannot support H-1B or STEM OPT candidates.

Worth knowing

Good:

• Pages are provided, so no unpaid time finding documents
• Clear, concrete task with a defined quality bar
• A route into reviewer work for accurate annotators
• Very active: 247 hires shown this month

Less good:

• $12.68 an hour for character-level precision on handwriting and dense layouts is modest pay
• Handling time is tracked, so slow careful work may count against you
• India-only
• Independent contractor; projects can be extended, shortened or ended early

About this listing

Posted by Mercor as a remote hourly contract with an India location requirement, read on 24 September 2026. The rate, workflow and requirements are the ad's own. See Mercor.