Semiqlassical
Introducing Toymodel
Complex computational workflows start with Toymodel. Toymodel is a human-inspectable computational workbench for industrial optimization, scientific machine learning, and physics-informed neural networks, with a built-in typed language, graph execution compiler, native tensor algebra, SQL-style tabular operations, and kernel-level acceleration.
About semiqlassical
Independent since 2024
Semiqlassical is an independent company, not funded by venture capital or big tech. We use that independence to build high‑precision physics ML and simulation software, held to a standard that meets or exceeds the best‑funded labs.
Independence lets us choose problems on their scientific merit and give precise work the time it needs. We would rather ship a tool that holds up under scrutiny than one that only demos well.
We started in 2024 with a focus on materials science and quantum materials, where a result only counts if the numbers hold. Those roots still shape everything we build.
Scientific computing
Our scientific computing work spans physics ML, simulation, and quantum‑inspired optimization: problems where an answer has to stay numerical, inspectable, and grounded in the physics.
Toymodel brings that work together. It compiles tensor algebra rather than looping in Python, which puts it in the performance class of C, Julia, and Rust, and its 128‑bit engine keeps long calculations from drifting. The compiler checks physical units before anything runs, and a full autodiff engine trains models in the same graph as your solvers and data.
Explore Toymodel →Slowdata
In real industries, the data that matters often lives on paper, in scans, and across legacy file stores. Slowdata lets people question those records in natural language and returns each answer with its evidence: the scan page, table cell, or image region it came from.
It runs on premises on our proprietary neuro‑symbolic engine, has been validated on more than 100 real‑world industrial data sets, and connects to your organization’s own AI systems. Slowdata makes the data verifiable; Toymodel makes the computation built on it inspectable.
Explore Slowdata →Folks at these organizations trust and use Toymodel
Founders
Avah Banerjee, PhD | Founder, CEO
Avah works across graph algorithms, high‑performance computing, and quantum computation. She has built compilers for quantum circuits, studied quantum walks, and designed resource‑efficient chaos generators. Her current research applies Markov chains and Monte Carlo methods, classical and quantum, to the modeling, the simulation, and the optimization of systems under real‑world constraints. She is former faculty, with a record of funded research and industry collaboration. At semiqlassical she directs the architecture, the numerics, and the mathematics behind our physics‑ML and quantum‑inspired tooling.
Emily Hsiang, PhD | Co-founder
Emily turns noisy signals into decisions. Trained in chemical engineering and neuroscience, she has built advanced optical imaging systems, developed ML pipelines that decode visual pathways, and shipped deep‑learning models that isolate signal from artifact. She works to three measures: operability, latency, and measurement, the three that carry a method from the laboratory into a running system. At semiqlassical she leads the data, the modeling, and the human factors of tools built for high‑stakes, everyday use.
Explore Toymodel
Start with Toymodel.
Toymodel is where the work begins. Open it, build a sector, and read every value on the way through.