Shubham User

Serves as a generalist AI trainer for advanced multimodal visual-content systems, performing annotation and reasoning task development, evaluation, and improvement.

Conducts adversarial AI evaluation and multilingual LLM training by designing culturally grounded prompts to test model reasoning, safety, and factual accuracy across Hindi, Hinglish, and English.

Trains Google Gemini using real-time speech and text data, focusing on factuality QA, speech annotation, and prompt construction within RLHF and SFT loops.

Performed bilingual English–Hindi annotation and linguistics work, including adversarial/red-teaming and evaluation of factuality, reasoning, cultural awareness, alignment, robustness, hallucinations, and bias.

Rated prompts and completions for relevance and factual correctness and contributed to multimodal visual QA, including response alignment and tone evaluation in English and Hindi.

Tuned Meta LLM responses for alignment and factual correctness using rubric-based QC and structured annotation tasks, including bilingual and multiturn evaluation and AI audio reviewing.

Contributed to translation QA, transcription alignment, multilingual dataset labeling, tone detection, and bilingual speech annotation.

Authored geography-based academic prompts and performed bilingual Hindi–English linguistic annotation, including tone and emotion tagging and transcription alignment with attention to regional nuance.

Handled factuality and audio reviewing for the Callisto project, running annotation QA to ensure consistency across speech/text quality checks.

Reviewed and rated Google Gemini outputs for truthfulness, fluency, and safety, using RaterHub and CrowdCompute for quality assessment.

Grade: 83/100

Achieved EF SET English certification at C2 proficiency (83/100).

Earned a professional certificate in Data Science.

Completed a Master's Degree in Commerce.