Vivek Sekhadia

• Co-led research on Retrieval-Augmented Generation and DPO-based actor-critic learning to reduce hallucinations in large language models, working alongside 3 peers and a mentor • Co-built the novel CREAM-RAG framework, sourced datasets, ran experiments across two models, and authored the experiments section of the paper • Outperformed three baselines: Pike-RAG, Self-RAG, and REAR-RAG on 5 datasets including RAGTruth (a rigorous hallucination corpus/dataset) across various metrics: F-1, EM, BERT, RAGAS Faithfulness, and RAGAS Correctness. • Research accepted to present at top AI/ML venues: AACL, AAAI, and NeurIPS/EurIPS • Earned a return offer from Algoverse with a 70% scholarship

• Co-led a Turing Test study under the mentorship of a Stanford graduate • Designed and executed a controlled experiment to evaluate an AI model’s ability to replicate human speech patterns • Collected and analyzed data from 50+ survey participants, producing empirical evidence that current AI systems struggle to reliably mimic human conversational behavior, identifying key deficiencies in linguistic nuance, coherence, and response authenticity • Presented findings to graduates from top schools like Harvard and MIT, incorporating expert feedback to strengthen experimental design and analytical rigor • Collaborated with 2 fellow students throughout the research process, coordinating data collection, analysis, and presentation of results • Designed and deployed a Turing Test simulation website, enabling real-time participant interaction •Designed and deployed a Turing Test simulation website featuring multiple difficulty levels, various AI models, and a live leaderboard for top scores

• Founded and scaled a free AI-powered writing style feedback tool that analyzes a user’s writing and provides a personalized summary of feedback based on their chosen style • Built the platform using React and JavaScript on the frontend and Node.js, Python, and SQL on the backend • Led a marketing team that scaled the platform through social media, growing to 240,000+ uses across 85+ countries with a 4.7/5 user rating • Received feedback from students to further improve the site and AI

• Founded and scaled an AI-powered research mentorship startup that guides students of all ages through the full research process. • Users choose their own topic to research and decide whether they want to learn or write a paper • Built and deployed a fine-tuned AI model that generates personalized research ideas and plans, paired with a secondary AI that acts as an on-demand research mentor students can consult at any point in their process • Built the platform using React and JavaScript on the frontend and Python and SQL on the backend • Grew to 15,500+ users across 75+ countries with over 1.2 million research ideas generated and a 4.8/5 user rating • Led a marketing team to scale the platform through social media