United States
Leading research on fundamental methods to train general reasoning models from unlabeled data. Developed and experimenting novel method grounded in Bayesian Inference and Inverse Reinforcement Learning.
Led research on Reinforcement Learning (RL) methods to transform Large Language Models (LLMs) into powerful sequential decision makers in complex reasoning tasks. Developed scalable RL algorithms to train specialized LLM agents for end-to-end software development.
Researched Graph Neural Networks (GNNs) and their capabilities on subgraph-level tasks. Developed novel pre-training and transfer learning frameworks to efficiently adapt pre-trained GNNs to subgraph-level tasks.
Designed and built a new production-ready context retrieval system for LLM prompt generation Significantly enhances LLM’s code-generation capabilities by providing relevant snippets of symbol definitions and references in user’s codebase Designed advanced preloading and asynchronous caching mechanisms to achieve low-latency retrieval