Pranava Mittal

Evaluates frontier model outputs for reasoning quality, instruction-following, safety compliance, and hallucination rate across generalist and specialist task domains. Builds adversarial prompts and edge cases, producing structured reports to refine alignment criteria.

Bilingually evaluates AI-generated text, audio, and personalization outputs in English and Hindi for naturalness, fluency, and cultural fit. Rates map personalization and search relevance for intent alignment and flags localization errors that impact product quality.

Contributed across 20+ RLHF task types (data annotation, audio evaluation, image analysis, reasoning, and creative writing), consistently ranking among top-rated contributors by task acceptance. Wrote comparative response rationales and fed preference data into RLHF/RLAIF training pipelines while identifying recurring failure patterns (hallucinations, unsafe outputs, instruction drift).

Grade: GPA 9.5 / 10 (Sem 1)

Activities and societies: AI trainer/evaluator contributions spanning data annotation, audio evaluation, image analysis, reasoning, and creative writing; focuses on detecting hallucinations, unsafe outputs, and instruction drift.

Pursuing a B.Tech in Computer Science at Chitkara University while contributing to AI model alignment through RLHF evaluation and data annotation across 20+ task types.