Deep-tech founder · Topaz Computing

Emma Leonhart

I'm a self-taught deep-tech founder. I build Topaz Computing, context management for people running a staff team of AI agents. The rest of this page is my other work: a programming language, a database, some tutorials, a few research projects. Separate things I made, not chapters of one argument.

◆ My company 💎 Topaz Computing Context management for agent teams.

Topaz is my company. It is building an ERP for agents rather than for humans: planning, records and issue tracking fused into one structured workspace that a team of AI agents works from, instead of a chat window. Delivery is a hosted box, a cloud container running persistent agents against your own repositories. It is aimed at solopreneurs running a staff team of agents, where keeping context straight across them is the thing that actually breaks.

topazcomputing.com →

01 — About

I'm a self-taught deep-tech founder. My company is Topaz Computing, described above. Everything further down this page is my other work, gathered here because I made it.

That list is a list. The projects are not instances of a single thesis and I am not going to pretend they are: Sutra is a programming language, Loka is a database, the tutorials are teaching tools, and the rest are their own things. Some of them fed into what I am building now and some of them did not.

I built all of it without a PhD, a lab, or an institution behind me, on the work itself: a language shipped to PyPI with a paper published on arXiv. Read the founder story →

The usual deep-tech founder has a doctorate and an easy fallback and hedges the bet accordingly. I have neither, and I'm not building one. I went in clear-eyed about exactly what I was giving up and committed anyway. That is the bet, and I am not hedging it.

02 — Sutra

📜 Sutra v1.1.0 · paper published
A geometric tensor programming language.

Every value is a tensor. Every operation is tensor arithmetic in a geometric space. There is no print, no if, no while: branches are continuous weighted blends and loops are geometric rotations. Programs compile to straight-line tensor operations, which makes them GPU-native and end-to-end differentiable by construction.

I wrote the reference compiler (PyTorch backend), the IntelliJ and VS Code plugins, the language specification, and the interactive demos, plus Loka, the bundled vector database. The name comes from the Sanskrit sūtra, the word used for Pāṇini’s grammar of Sanskrit, the earliest known formal grammar of any language.

Sutra is closed at v1.1.0 and is on PyPI as sutra-dev. The paper is published on arXiv (2605.20919, cs.LG). It is not what Topaz sells, and it is not the pitch. It is the thing I built before it.

03 — Other things I've built

Projects
Loka, QueryKey, Reservoir Agent, cleanvibe, Latent Space Cartography, the alignment work, and the vibecoding tutorial. Each on its own emmaleonhart.com subdomain.
Interactive ML Tutorials
16 hands-on visualizers covering vector math, neural networks, training dynamics, and modern architectures. Drag, click, change the inputs, watch the math.
Loka Theory
Eight interactive explainers for the database internals: HNSW in RDF, subgraph SIMD indexing, SPARQL exit conditions, and how graph and vector databases differ.
Research
Papers and research artifacts, including Sutra (published) and Latent Space Cartography (on hold at arXiv).

04 — Links

emmaleonhart.com