Native-speaker Indic eval for the languages IndicVisionBench cares about

#2
by pferg42 - opened

Hi Krutrim team,

Congrats on IndicVisionBench landing at ICLR 2026. Building your own culturally grounded, multilingual benchmark for the Indian context says the quiet part out loud: the hard part of Indic evaluation is not raw model scale, it is native-speaker judgment of fluency, preference, and cultural correctness across a lot of languages. On top of Krutrim-2 and Krutrim-Translate spanning Bengali, Hindi, Kannada, Marathi, Malayalam, Gujarati, Telugu, Punjabi, and Tamil, that judgment layer is exactly the part that still needs people.

That is what we run. Boltwork is a human-data network of native speakers in long-tail languages, with a live Hindi panel now and more Indic languages available from the same crowd. Quality is built into the pipeline: hidden gold checks, multi-judge consensus across distinct workers, per-worker trust scores, and a written rationale plus an error-category tag and provenance on every judgment, so the labels are auditable, not a bare score.

The offer is free: send 100 to 300 of your own items (preference pairs, output ratings, or fluency judgments) in any of your Indic languages, and we return labeled JSONL in about a week with full QC stats. On our pilot set we run about 92% gold accuracy and 95% inter-annotator agreement. No cost, no commitment. The cleanest read is on a couple of your scarcest languages.

One operational detail that helps us reach and keep native speakers across India: we pay them instantly in Bitcoin over Lightning, which sidesteps the payout and KYC friction that usually breaks the unit economics at the long tail.

Methodology and deliverable format: https://lightningfaucet.com/boltwork/data/. Reply here or at partnerships@team.lightningfaucet.com and we will send a redacted sample record.

Worth 15 minutes?

Boltwork Team

Sign up or log in to comment