Now
I’m building Semiqlassical, the startup building the computational and trust layer for semiclassical physics, which I co-founded with Dr. Avah Banerjee.
It grows out of the same passion that pulled me into science in the first place: solving hard problems without losing the reasoning that makes the result meaningful. I care deeply about the way AI can hide the work and flatten education into answer-getting without the thinking process. I built biological brain simulations and machine learning models to understand how information transforms through neural circuits; now I’m building tools for the next generation of physics simulation and ML in semiclassical physics, with the process kept visible enough to inspect, learn from, and trust.
Attryx
Attryx is our first product at Semiqlassical: a human-inspectable computational workbench for frontier research, from protein dynamics to quantum circuit simulation.
We are building it for workflows where the path from input to result matters as much as the answer. Code, AI-assisted steps, diagnostics, tables, and results should remain visible enough for people to examine, share, and build on top of.
That matters especially for the next generation of researchers, from college students and undergraduates to graduate students and early technical teams, who need AI to expand their thinking without hiding the process that makes scientific judgment possible.
Visit Attryx.aiAbout Me
I have been fascinated by the brain for as long as I have been building things: not only because it is efficient, but because it computes under constraints that today's large language models do not yet capture. Brains learn from limited data, act through bodies, carry fear, uncertainty, and mortality, and turn biological signals into decisions that become behavior.
At Semiqlassical, that question has become a product and research agenda. I co-founded the company with Dr. Avah Banerjee, and we are building Attryx as a scientific computing platform for frontier research, from protein dynamics to quantum circuit simulation and related technical workflows: a place where AI can help researchers reason, simulate, and build, while keeping code, assumptions, intermediate steps, diagnostics, and results visible enough for human examination.
My founder thesis is not just "human in the loop." As technology becomes more powerful, human judgment is the value-bearing part of the system. The harder question is how we determine, inspect, and preserve those values as AI makes results cheaper and faster. If the process disappears into a black box, we lose the evidence trail that lets science, society, and individual people decide what is true, what matters, and how to move forward.
My scientific path has followed the same thread from computation to biology and back. At Washington University in St. Louis, I earned my PhD in Neuroscience with Dr. Daniel Kerschensteiner, studying how retinal circuits compute. My work used multiphoton calcium imaging, multi-electrode array recordings, large-scale data processing, and neural-network modeling to study subcellular processing, parallel visual pathways, efficient coding, and natural-scene representation in mouse and human retina.
Recently, I developed a query-based neural activity prediction model (qNAP) designed to generalize across neurons by disentangling descriptors such as receptive-field location and cell-type structure. I am interested in foundation-style models for biological visual pathways: models that do not merely fit one experiment, but help us ask what kinds of structure, data, and computation are needed to explain living circuits.
Before graduate school, I worked in cognitive neuroscience and human epilepsy research as a Visiting Scholar in the Parvizi Lab at Stanford University and as a Research Assistant at the Institute of Cognitive Neuroscience, National Central University in Taiwan. Those projects shaped how I think about decision-making, electrical connectivity, deep-brain recordings, and the way internal brain dynamics become visible as behavior.
I began in Chemical Engineering at National Taiwan University, where my undergraduate thesis focused on photoactive catalytic reactions. Even then, I was drawn to the idea of building artificial systems that could help us understand the human brain. Today, I see AI as only the beginning of that effort. We will need more diverse computational paradigms, including simulation on advanced quantum and neuromorphic hardware, to understand intelligence, biological constraints, and ourselves with the depth this moment demands.
Fiction and Art
I also write science fiction. One question I return to often is what happens when a single physical principle changes: how a world reorganizes itself, what kinds of lives become possible, and what story that world ends up telling about itself.
That instinct is closely tied to the way I experience art. I like staying with a work longer than most people do, looking for the formal decisions, emotional residue, and small clues the artist left behind. In both fiction and art, I am drawn to the way a world reveals its rules indirectly, and to how meaning accumulates through structure, constraint, and sustained attention.
Academic Archive
For my academic research archive, publications, and earlier work under the name Jen-Chun Hsiang, visit JenChunHsiang.com.
Research Background
These selected papers reflect the research path behind my current work. For the fuller publication record, visit Google Scholar, and for the archive, visit JenChunHsiang.com.
- Explicit disentanglement of neural descriptive factors using the query-based neural activity prediction (qNAP) model. COSYNE, 2025. [Poster]
- Distributed feature representations of natural stimuli across parallel retinal pathways. Nature Communications, 2024. [Paper]
- Dendritic and parallel processing of visual threats in the retina control defensive responses. Science Advances, 2020. [Paper]
- Efficient Coding by Midget and Parasol Ganglion Cells in the Human Retina. Neuron, 2020. [Paper]
- A systematic study of stereotypy in epileptic seizures versus psychogenic seizure-like events. Epilepsy & Behavior, 2019. [Paper]
- Local processing in neurites of VGluT3-expressing amacrine cells differentially organizes visual information. eLife, 2017. [Paper]
- Distinct patterns of temporal and directional connectivity among intrinsic networks in the human brain. Journal of Neuroscience, 2017. [Paper]