Founder & CEO, Sober Agentic Infrastructure, Inc. · Operational Sobriety for the Agentic Era. Founder & Creator, VAREK · AI Pipeline Programming Language.

I am a physician, systems engineer, and language designer dedicated to building the foundational trust layer for autonomous AI. As AI agents move from "chatbots" to "executors," the risk of statistical intoxication (hallucinations) creates massive enterprise liability — and the fragmentation of AI tooling compounds every failure surface.

I build two things to fix that.

varek-lang.org · github.com/kwdoug63/varek · MIT License

VAREK is an open source programming language designed from the ground up for AI/ML pipeline engineering. It replaces the Python + YAML + JSON Schema + Bash fragmentation that plagues production AI systems with a single typed, compiled language where pipelines are first-class citizens.

• Statically typed — Hindley-Milner inference extended with tensor shape tracking
• LLVM-compiled — 10–40x faster than CPython for compute-heavy pipeline workloads
• Pipeline declarations — type-checked at compile time, not discovered at runtime
• Full interop — import python::numpy, import python::torch — no rewrites required
• v1.0 stable — 659 tests passing · 7-module standard library · 261 functions · package manager

VAREK is the language that Sober Agentic Infrastructure runs on. Compile-time pipeline verification is deterministic AI safety at the language layer.

soberagents.ai

We build model-agnostic, neuro-symbolic guardrails that sit between AI agents and critical infrastructure.

• Deterministic Safety: Using the Z3 SMT Solver to mathematically block rogue actions in <50ms.
• The Shadow World: Our proprietary simulation engine that proves agent plans are safe before they execute.
• Liability Mitigation: Closing the "Trust Gap" that prevents Fortune 500s from deploying autonomous agents at scale.

My approach to AI safety is not theoretical; it is forged in environments where failure is not an option:

• Medical Precision: Doctor of Medicine (MD), USC Keck School of Medicine. I understand systemic failure and the necessity of "clinical" guardrails in autonomous systems.
• Language Design: Built VAREK from first principles — lexer, parser, type system, LLVM backend, standard library, package manager. 659 tests. 13,006 lines. One year.
• Hardware Innovation: Lead Author of US Patent 12,037,144 B1 (Space Railway). I build systems that rely on the laws of physics and logic, not probability.
• Software Innovation: Lead Author of US Provisional Patent 64/006,104 • Filed Mar 15, 2026 (Deterministic Safety Guardrails and Outcome Verification for Agentic Systems). I build systems that rely on the laws of math and cryptography, not probabilistic guesswork.

• VAREK: Python (reference implementation) · LLVM IR via libLLVM-20 · Hindley-Milner type system
• Sober Agentic Core: Python 3.11+ · Rust (Logic Kernel) · FastAPI · Docker
• Formal Methods: Z3 SMT Solver · First-Order Logic (FOL) predicates · Symbolic Verification
• Predictive Math: NumPy/SciPy for latent-space bounding and action-validation

• VAREK: varek-lang.org
• Sober Agentic: soberagents.ai
• LinkedIn: linkedin.com/in/kwdoug63
• Email: kenneth.douglas@soberagents.ai

"We don't need smarter models; we need soberer infrastructure."