Scientist working at the interface of cell therapy, computational biology, and AI-enabled drug discovery.
I bridge experimental and computational discovery. I build AI-powered platforms, deep learning models, and multi-omics and immune-repertoire pipelines that turn single-cell and NGS data into translational insight for TCR and cell therapy discovery, and I test what they predict with my own hands at the bench.
What excites me most right now: AI agents that take on real problems in biomedicine and drug discovery.
Models and software I’ve built to shorten the distance from sequencer to decision.
Nine agent skills behind one entry point that turn a vague enterprise AI request into a project plan you can defend: evidence first, human sign-off last.
An AI agent that lets bench scientists run TCR and antibody sequencing pipelines on a shared HPC cluster by describing what they need in plain language.
A protein language model that ranks T-cell receptors by the disease they likely recognize, so the lab screens a short list instead of a whole repertoire.
Read the case study: TCR Pathology ClassifierSource on GitHub
Twenty-one open-source agent skills that walk a project from idea to release — research, PRD, spec, design, test-driven build, and retrospective — with real stop points.
Read the case study: Full-Chain Development SkillsSource on GitHub
Following two cell therapies through tissues and time to learn why one of them does better in solid tumors — and which checkpoint each one needs.
How immune cells sense, signal, and kill: from nanoscale receptor imaging to engineered cell therapies.
Signal Transduction and Targeted Therapy 2026
We followed conventional CAR-T cells and stem cell-derived CAR-NKT cells across tissues and time in solid-tumor models. CAR-NKT cells homed into tumors better and persisted longer,…
We followed conventional CAR-T cells and stem cell-derived CAR-NKT cells across tissues and time in solid-tumor models. CAR-NKT cells homed into tumors better and persisted longer, and the two products responded to different checkpoint blockades: TIGIT for CAR-T, CD96 for CAR-NKT.
Graphic: Conceptual schematic linking spatiotemporal single-cell profiling to matched checkpoint therapies: CAR-T with anti-TIGIT and CAR-NKT with anti-CD96, followed by experimental validation in solid-tumor models.
Cell 2025
Tumor-infiltrating CD8 T cells switch on the serotonin transporter SERT, which acts like an immune checkpoint. Widely prescribed antidepressants (SSRIs) that block it boosted anti-tumor T…
Tumor-infiltrating CD8 T cells switch on the serotonin transporter SERT, which acts like an immune checkpoint. Widely prescribed antidepressants (SSRIs) that block it boosted anti-tumor T cells and worked together with PD-1 blockade in mouse models.
Graphic: Bioinformatics and functional validation connect SERT blockade to enhanced CD8 T-cell activity: single-cell and human tumor analyses support the serotonin mechanism, while fluoxetine plus anti-PD-1 improves tumor control in mice.
Nature Biotechnology 2025
A feeder-free, clinically guided process turns blood stem and progenitor cells into IL-15-enhanced allogeneic CAR-NKT cells at high yield and purity, with anti-tumor activity across several…
A feeder-free, clinically guided process turns blood stem and progenitor cells into IL-15-enhanced allogeneic CAR-NKT cells at high yield and purity, with anti-tumor activity across several cancer models.
Graphic: Cord-blood stem and progenitor cells become allogeneic CAR-NKT cells through a six-week culture. Single-cell and methylation analyses explain effector and memory states and low immune-marker expression, supported by functional tests and preclinical myeloma results.
Science Advances 2020
We built nanoarrays that mimic the contact zone between a macrophage and a microbe. Moving ligand clusters closer together brought receptor clusters together and increased inflammatory…
We built nanoarrays that mimic the contact zone between a macrophage and a microbe. Moving ligand clusters closer together brought receptor clusters together and increased inflammatory signaling — up to an intrinsic ceiling.
Graphic: A macrophage rests on an artificial phagocytic synapse of Pam3-coated gold nanodisks and IgG-coated glass. A schematic shows the spacing limit of TLR1/2 clusters and the TNF-alpha response plateau at close ligand spacing.
Explainers, short videos, and working notes on statistics, machine learning, and immunology.
Video, 3:35
A 3.5-minute narrated video and a complete tutorial on using DESeq2 (PyDESeq2) to compare V-gene usage between CAR-T and control T cells: why raw percentages mislead, the five steps inside DESeq2, a…
Video, 5:35
A 5-minute narrated video and an illustrated walkthrough of the Kaplan–Meier survival curve: time to an event, censoring, why counting fails, the step-by-step calculation, medians and confidence bands, the log-rank test, the…
Video, 9:34
A 10-minute narrated video and an illustrated walkthrough of the four stages behind GPT and ChatGPT: pretraining, supervised fine-tuning, a reward model trained on human rankings, and RLHF with PPO. Every loss…
From physics to the bench to code, with immunology as the constant.
2023 – Present
TCR discovery, immune-repertoire and single-cell pipelines, deep learning for biology discovery, AI tooling for NGS analysis.
2022 – 2023
Cell therapy bioinformatics.
2018 – 2022
Computational imaging of immune receptors.
2013 – 2017
Munich, Germany
2011 – 2012
Nottingham, United Kingdom
2006 – 2011
Wuhan, China
Half of it runs on a cluster, half of it in a biosafety cabinet.
Multi-agent analytics workflows, with human approval, identity isolation and audit logging built in.
Reusable, tested, FAIR-aligned pipelines and production-grade web apps.
From classic ML to fine-tuned foundation models, validated against held-out functional readouts.
Patient-derived oncology and immune-disease datasets, from QC and integration to mechanism.
Primary and stem cell-derived immune cells, engineered and tested for function.
The constructs and edits behind engineered cells.
Seeing receptors, cells and animals, from single molecules up.
Models built at the computer, tested at the bench, and tools that bring the two together.
I’m glad to talk about collaborations, invited talks, and roles in computational immunology, AI/ML, and clinical data science.
