SkyRL
Contributing to SkyRL, an open-source modular reinforcement learning framework for LLM post-training, with work across rollout generation, Megatron parallelism, vLLM/SGLang integration, GPU CI, and distributed RL training infrastructure.
Projects
Selected systems and research work.
AFib Risk Prediction
Large-scale AFib risk prediction and patient trajectory modeling using longitudinal EHR data, deep learning, and medical foundation models at BAIR and Computational Precision Health.
MicroML
Built a lightweight deep learning framework in modern C++ with automatic differentiation, SIMD-optimized tensor kernels, and end-to-end neural network training infrastructure.
Introspect AI
Built a 3D semantic visualization system for exploring AI conversation histories using embeddings, UMAP, clustering, and interactive WebGL interfaces.
Agile Agents
Built a multi-agent software engineering framework on top of SWE-Agent to study how collaborative LLM teams perform on real GitHub engineering tasks.
UniTrial
Built a multimodal RAG system for clinical trial recruitment using EHR retrieval, medical embeddings, and live ClinicalTrials.gov data.
Neurohack C3
Built a computational genetics and drug discovery pipeline for identifying ALS-linked genetic variants, causal disease mechanisms, and druggable targets.
Project BlueSky
Built a multimodal mental health monitoring system combining facial expression analysis, speech emotion modeling, ASR, and NLP-based conversation understanding.
PYR Health
Designed a low-cost microfluidic blood monitoring device for remote CBC testing and post-chemotherapy patient care.