Tauska Wang

I’m a machine learning engineer focused on post-training large language models and high-integrity data pipelines—especially where scientific rigor and real-world impact intersect. 

My work spans agentic systems, reward modeling (RLAIF/RLHF), and evaluation at scale, with a track record of building reproducible tooling and shipping measurable improvements.

Recent focus areas:

LLM post-training & evaluation: Scaled rubric-based reward datasets to the million-pair range; trained summary/CoT reward models; delivered a ~5% lift on a health benchmark.

Agentic systems & tooling: Built Pantheon-CLI and an LLM query-router; delivered end-to-end single-cell analysis agents; added cross-provider API compatibility; surpassed a general SWE-agent CLI on biomedical tasks.

Scientific/health AI & data engineering: Parallelized BKMR on k8s; vectorized 30 years of NHANES; developed TCGA/GEO pipelines; created a mixture-of-experts framework for COPD metal-risk assessment; integrated wet-lab and dry-lab workflows.

Quality & evidence: Author on peer-reviewed studies; reviewer for leading journals and ICLR; co-inventor on patents spanning transformer-based multimodal medical AI and biotechnology.

Tooling & stack: Python, PyTorch, PyG, SQL; GCP/Azure; open-source contributions (PantheonOS/Agentic Data Science, OmicVerse, RAG Web UI, AstrBot). Currently pursuing advanced studies bridging medical science and computer science, and I thrive on projects where careful ablations, clear metrics, and tough benchmarks drive decisions.

If you’re tackling LLM post-training, agentic workflows, or scientific/health data at production scale, I’m interested in collaborating on systems that move from promising results to verifiable outcomes.

My research integrates reproductive endocrinology, metabolism, and artificial intelligence to uncover mechanisms of endocrine–metabolic disorders such as polycystic ovary syndrome (PCOS). I develop AI-driven frameworks for multi-omics integration and single-cell data analysis to reveal cross-organ communication among reproductive and metabolic tissues. I also explore large language models (LLMs) for biomedical text mining, literature synthesis, and clinical decision support. By combining systems biology with machine learning, my goal is to advance precision medicine and digital health approaches that bridge molecular discoveries with real-world clinical data, ultimately improving diagnosis, prevention, and treatment of reproductive–metabolic diseases.