Anirudha Ramesh
I’m a Research Engineer at InstaDeep, working on LLM post-training, Deep Reinforcement Learning, and AI for Science. I’m building DeepPCB, the leading PCB routing and placement product, and contributed to Laila, the world’s first AI Lab Assistant (covered in FT, Benzinga, and more). I also mentored research pushing the frontier in long-range genomics, enabling inference of ultra-long sequences >1Mbp on a single GPU with zero-shot generalization to sequences 100× longer than seen during training (ICLR 2025 LMRL Website & Spotlight Talk).
I’m currently also working on improving compute-aware inference strategies and advancing AI safety via weak-to-strong generalization with a focus on effective oversight as models get stronger.
Background
I completed my Masters in Robotics at Carnegie Mellon University, advised by Jeff Schneider and Christoph Mertz. I designed and deployed a perception system for 24/7 off-road autonomy using multi-spectral inputs, and introduced a domain adaptation framework that outperformed existing methods by +40% mIoU (Thesis Talk, Paper).
Previously, I interned at Adobe’s Media and Data Science Research Lab, where I discovered and solved biases in Few-Shot Segmentation datasets and methods, improving performance by ~5% mIoU (NeurIPS 2021).
I studied Computer Science at IIIT Hyderabad (CGPA: 9.32/10.0, Dean’s Research Award), working with Professor Madhava Krishna on monocular multibody SLAM for autonomous driving, achieving 3× smaller tracking error than previous state-of-the-art (IEEE IV 2020, Video, IV 2021, VISAPP 2021).
Research Interests
- Agentic AI systems
- Deep Reinforcement Learning for real-world optimization (PCB design, LLM post-training)
- AI for Science (genomics, drug discovery, materials)
- Computer vision in challenging domains (multi-spectral, off-road)
- Domain adaptation and few-shot learning
- Robotic learning