कर्मसु कौशल्यम् | "skill in actions"
Hey! I am Uday Ramesh Phalak.
This space documents my applied research at the intersection of advanced AI, human-centric design, and my primary focus: AI Safety. My research centers on developing machine learning systems that are robust, effective, and aligned with human needs. Building on my early work in adversarial ML defense (2017–2019), my primary goal is to ensure that as AI advances, it remains interpretable and truly safe.
Research Focus Areas:
- Mechanistic Interpretability & AI Safety: Decoding the black box. I evaluate and interpret how models operate to ensure they can be safely applied to high-stakes domains such as health, physics, and others.
- Alternative Architectures & Neurosymbolic AI: Pushing beyond standard models. I am experimenting with scaling reasoning via RL, transformer-free and energy-based models, and currently building a custom Sanskrit Grammar LLM driven by Neurosymbolic AI.
- Generative UX and Recommendations: I research generative recommendation systems to enhance the intuitiveness and fluidity of human-AI interactions.
- Scalable Multi-Agent Systems & Optimization: I bridge research and machine learning engineering by focusing on scalable multi-agent systems and optimization. My work includes reinforcement learning training, rigorous benchmarking, and scaling multi-agent architectures. I optimize system performance to ensure complex AI models move efficiently from research to production.
Subscribe to my blog via email or RSS feed.
/. Latest Weekly Updates
- LTX-2: Decoupling Audio-Video Streams for 20x Efficiency
- Why we built agentic skills for rust language ?
- Activation Capping: The Digital Lobotomy of AI
- Beyond "Chat Completions": Open Responses, the Universal Standard for Agentic AI
- Universal Commerce Protocol : Solving the "N x N" Integration Bottleneck