AI that works like software, not chat — code that builds code, backed by 20 years of shipping production systems.
We build developer leverage: frameworks, DSLs, optimizers, agentic pipelines, static analyzers, and FFI bindings — in typed ecosystems (Kotlin/Java, Scala, Rust, TypeScript), where type discipline makes tooling safer and faster to build. We treat AI as disciplined engineering leverage, directed by human judgment the way a profiler or debugger is.
Everything here is independently verifiable: public code, a published DOI, live hosted products, and 57k+ downloads — no vouching required.
| Project | What it is |
|---|---|
| Cognotik | AI productivity platform — hosted SaaS + open-source local-first app. AWS ECS/Fargate, Terraform, OAuth, Stripe billing. 8+ model providers, agentic workflows, ~95% AI-generated with human review. → cognotik.com |
| QQN Optimizer | Novel Quadratic Quasi-Newton optimizer — formal convergence proofs, 72.6% benchmark win rate. Rust framework + production JAX/Optax port. → DOI |
| Mathematical Explorations | Original interactive in-browser math research labs & essays. Zero build step, deployed via Terraform. → math.cognotik.com |
| Arcade of Life | Zero-dependency PWA — arcade games on Conway's Game of Life + 50 CA rulesets, CPU/WebGL2 backend, hashlife. → aol.cognotik.com |
| MindsEye | Deep learning framework built from scratch in Java with custom CUDA/CuDNN — before TensorFlow's first release. |
Kotlin · Java · Rust · TypeScript · Scala · Python (JAX/Optax) · C/C++/CUDA
AWS · Terraform · Docker · Kubernetes · Spark · PostgreSQL
LLM Orchestration · Agentic Workflows · RAG · Developer Tooling · Numerical Optimization
- 57,000+ downloads of an early-market JetBrains AI plugin (predating ChatGPT)
- 72.6% benchmark win rate for the QQN optimizer (published, DOI-backed)
- <5ms TP95 at ~10k TPS real-time systems experience
- ~95% AI-generated codebase (Cognotik) with human review — disciplined AI leverage in practice
- Code that builds code. The highest-leverage software is the software that writes, refactors, and verifies other software.
- AI under human judgment. LLMs are a force multiplier we design and direct — auditable, human-in-the-loop, not opaque autonomy.
- Verifiable over vouched. Public code, published proofs, and live products beat claims every time.
- Build to understand. We perturb rigid structures with one irrational twist and study the rich behavior that emerges.
- 🌐 Website — simiacrypt.us
- ✍️ Blog — blog.simiacrypt.us
- 🤖 Cognotik — cognotik.com
- 📐 Math Labs — math.cognotik.com
- 📄 QQN Paper — DOI: 10.13140/RG.2.2.15200.19206
- 📺 YouTube — @Cognotik
- 💼 LinkedIn — andrewcharneski
- 📧 Email — andrew@simiacryptus.com
