Sylendran Arunagiri

Sylendran Arunagiri is a technical product marketing manager for NVIDIA NeMo, specializing in agentic and generative AI. He focuses on making enterprise AI agents more reliable and economical by continuously learning from production data instead of depending on increasingly large models.

Before joining NVIDIA, Arunagiri worked at Freshworks and developed hands-on experience in technical product management and AI development. He studied at Carnegie Mellon University, the Indian Institute of Management Bangalore, and the National Institute of Technology Karnataka, Surathkal. His experimental AgentWeb project coordinated specialized product-management, development, design, and research agents to create personalized portfolio websites.

• Production data flywheels: Arunagiri combines user feedback, inference logs, curated training examples, human review, evaluation, and model deployment into a continuous improvement cycle. Human reviewers distinguish routing failures from retrieval problems and other errors before adapting models.

• Smaller models tuned for specific tasks: For NVIDIA’s internal employee-support agent NVinfo, he examined how employee feedback and expert review improved routing among specialized assistants. A fine-tuned 8-billion-parameter Llama model matched a 70-billion-parameter model’s routing accuracy in that evaluation, illustrating how task-specific optimization can reduce latency and inference costs. The NVIDIA Data Flywheel Blueprint extends this approach through automated evaluation and model distillation.

• Reinforcement learning with verifiable rewards: His writing on agent reinforcement learning emphasizes checkable outcomes, dependable reward functions, failure analysis, and continuous evaluation.

• Solver-backed agent skills: His supply-chain optimization work pairs language-model interpretation with NVIDIA cuOpt, assigning production-planning and inventory calculations to specialized mathematical optimization tools.