Khushi m mishra

I've delivered 98%+ accurate Hindi and English transcriptions across varied audio qualities, accents, and dialects, including speaker diarization, timestamping, editing, proofreading, and contextual annotation for AI training and enterprise projects.

I evaluate LLM and voice-model outputs for accuracy, reasoning, safety, pronunciation, prosody, pacing, and naturalness. I create prompts, benchmark model responses, perform preference ranking for RLHF, fact-check claims, and identify adversarial edge cases to strengthen multilingual AI systems.

My work includes Hindi-English localization, translation quality review using MQM/LQA frameworks, bilingual glossary development, safety-data labeling, and red-teaming. I use Express Scribe, Otter.ai, Label Studio, CVAT, Audacity, Adobe Audition, and AI chat interfaces while protecting data confidentiality and meeting demanding quality targets.

Alongside language-data work, I've taught Hindi, Sanskrit, and related subjects in college and school settings, developed lesson plans, mentored learners, and supported junior transcribers and reviewers on QA practices.