About Avalisa

Independent Researcher / ElmWater AI

avalisa@elmwaterai.com San Francisco, CA

I am an AI researcher focused on reinforcement learning, policy optimization, and large-scale learning systems.

My work spans theory, controlled empirical study, and production-scale learning systems, including RL algorithms, LLM post-training, evaluation systems, and multi-agent deployment.

Experience

Independent AI Researcher

2025 - Present

  • Developing a unified bias-variance analysis of GRPO-style operators for policy optimization.
  • Studying how filtering, token weighting, and advantage normalization affect nonconvex optimization through bias and covariance channels.

Research Collaborator at ElmWater AI Lab

Jun 2025 - Present

  • Conducted RL post-training on 32B to 70B language models in simulated decision environments.
  • Designed reproducible LLM evaluation across 12 benchmarks with experiment tracking and systematic comparisons.
  • Ran 100+ controlled experiments comparing 5 policy optimization methods across 8 tasks and multiple random seeds.

AI Engineer at Travelers Insurance Company

Jul 2023 - Mar 2025

  • Architected a distributed multi-agent LLM system for document analysis and task processing, scaling to 1,000+ complex cases daily.
  • Fine tuned large language models on private domain data for production-grade internal use.

Co-Founder and Lead Software Engineer at NearMeNow

Jan 2022 - May 2023

  • Co-founded a consumer technology startup building a distributed real-time location platform for nearby activities and events.