Varsha Kuppur Rajendra

About Varsha Kuppur Rajendra

Applied Scientist | LLMs | NLP | AI for Speech & Text | Debiasing & Model OptimizationI specialize in Large Language Models (LLMs), NLP, and AI-driven speech & text applications, focusing on data augmentation, debiasing, and model robustness. My work spans Text-to-SQL, sentiment analysis, scalable data extraction, and medical AI applications. I have published papers in AI research and published multiple patents in NLP robustness, Text-to-SQL, and AI optimization.At Oracle AI, I have:Enhanced NL2SQL models, improving multi-turn follow-ups, schema understanding, and SQL prediction using logical business rules.Applied Self-Refinement Training (SRT), Reinforcement Learning from Human Feedback (RLHF), and Direct Preference Optimization (DPO) to improve model robustness and alignment.Experimented with contrastive learning, invariance losses, and other optimization techniques to improve generalization and response quality.Leveraged Pattern-Exploiting Training (PET) for semi-supervised data augmentation, enhancing model adaptability with task-specific prompts.Developed large-scale multilingual data pipelines, extracting 54M+ reviews from Common Crawl across English, Spanish, French, and Portuguese, and designed behavioral checklist tests for Spanish, improving model robustness and evaluation.Advanced medical AI applications, optimizing speech-to-text and text-to-speech models for clinical use.With a Master’s in Computational Data Science from Carnegie Mellon University and multiple patents in NLP robustness, Text-to-SQL, and AI optimization, I am passionate about pushing the boundaries of trustworthy and scalable AI.

Experience

1. Sr Machine Learning Engineer

   Oracle

   Jul 2025 — Present · Seattle, WA, US

   Agentic AI Platform for Region Build Analysis– Led the design and productionization of a multi-agent GenAI system and RAG architectureover large-scale logs and tabular data from ∼ 50k failures across 150+ regions to extractoperational insights for cloud deployments.– Built a deterministic evaluation and validation pipeline for LLM-generated insights, enablingaccurate and reliable reporting despite the stochastic nature of LLM outputs.– Productionized AI services as containerized CronJobs, implementing parallelized data extrac-tion and generation pipelines across multiple cloud regions.Technology: LangChain, CrewAI, Terraform, Kubernetes (OKE)

Education

• Carnegie Mellon University

  Master of Computational Data Science

  2019 — 2020

• RV College Of Engineering

  Bachelor of Engineering - BE, Computer Science

  2013 — 2017