Rahul Oberoi

- Gold Level Consultant

- Enhanced the multimodal accuracy of Grok 4.0, Grok 4.1, and Imagine 1.0 by identifying and addressing systematic output errors within the data team. - Conducted root-cause analysis on vision and audio model failures, curating high-quality datasets to directly mitigate hallucinations and edge-case mistakes.

- Engineered an end-to-end KG-RAG system for Swiggy Instamart that automates Root Cause Analysis (RCA) on P&L metrics, utilizing cosine similarity for high-precision noise filtering. - Transitioned early from this internship to accept an offer at xAI.

Guide: Dr. Ganesh Bagler - Developed Named Entity Recognition (NER) models to extract structured information from unstructured recipe text (ingredients, quantities, units, states). - Worked with a novel manually annotated recipe corpus, contributing to a new dataset for computational food research. - Implemented models using spaCy and FLAIR, achieving 92.78% macro F1 on the test set without contextual embeddings.

- Evaluated and improved LLM chatbot responses to enhance accuracy, reasoning, and multilingual performance. - Contributed to prompt fine-tuning and training data quality improvements, supporting the development of a multilingual conversational AI system. - Became a Top Rated freelancer in less than a month with a 100% Job Success Score

Guide: Dr. Vinayak Abrol - Successfully identified critical features crucial for detecting synthetic speeches. - Conducted in-depth analysis to discern specific frequencies significantly influencing deepfake identification. - Carried out structured quantization. - Managed to reduce the training time to 15 minutes per epoch from 2 and a half hours while maintaining a comparable EER (Equal Error Rate)

Guide: Dr. Sachin Gupta