Shivani Singh

NVIDIA Robotics Project. Engineer structured training datasets for robotics AI pipelines in collaboration with client engineering teams. Implement preprocessing, transformation, and validation workflows using Python in Linux environments to produce model-ready data. Improved dataset validation consistency by ~4-5% through systematic error analysis and quality controls. Reduced downstream rework by enforcing strict alignment with training specifications. Support iterative training cycles by maintaining scalable and reliable data inputs within enterprise-grade workflows.

Fine-tuned Large Language Models using Reinforcement Learning via Human Feedback (RLHF). Designed high-quality prompts to improve reasoning depth, coherence, and alignment. Evaluated and benchmarked model outputs against structured quality metrics.

Increased NLP classification accuracy by 23% through feature engineering and targeted error analysis. Developed Python-based ML inference APIs achieving ~80 ms latency. Designed RESTful services enabling reproducible and scalable model deployment. Applied linear and integer programming techniques for structured optimization tasks.