Summary
Manqing Liu is a research-driven software engineer and PhD candidate specializing in causal machine learning and AI safety, with eight years of experience bridging statistical health research and scalable ML systems. Currently a Teaching Fellow at Harvard, she teaches high-performance parallel programming (OpenMP, MPI, CUDA) and mentors projects optimizing LLM inference (vLLM, KV-cache, continuous batching, tensor parallelism). At Geodesic Research and the Cambridge AI Safety Hub she built evaluation frameworks and multi-GPU pipelines to detect pathological Chain-of-Thought behaviors like post-hoc rationalization and steganographic encoding. Her background in biostatistics and epidemiology informs a rigorous, measurement-first approach to model evaluation and alignment. Known among peers as a Junior Fellow at Harvard, she combines deep academic training with hands-on engineering—shipping open-source tooling for multi-GPU training and inference optimization. Colleagues note she often frames safety problems through causal questions, revealing subtle failure modes that standard benchmarks miss.
8 years of coding experience
3 years of employment as a software developer
Doctor of Philosophy - PhD, Causal Machine Learning, Doctor of Philosophy - PhD, Causal Machine Learning at Harvard University
Master of Health Science (MHS), Epidemiology, Master of Health Science (MHS), Epidemiology at Johns Hopkins Bloomberg School of Public Health
Post-Baccalaureate Studies, Post-Baccalaureate Studies at University of Pennsylvania