What the work looked like

Four engagements, including the parts that were harder than expected.

Speech data for a voice assistant in eleven Indian languages

Eleven languages, district-level quotas and far-field capture, delivered in fourteen weeks.

A deliberately messy layout mix, with redaction verified before any annotator saw a page.

Written rationales on every comparison, with agreement published per language.

Frame-level state annotation with agreed onset and offset, delivered against a fixed homologation date.

Bring it to a scoping call and we will tell you how close the comparison really is.