I build across the stack—from RTL and operating systems to ML runtimes and reproducible experiment platforms. The common thread is simple: make the mechanism inspectable, then make the claim testable.

logic gates  →  processors  →  runtimes  →  learning systems  →  evidence

Five-stage RV32IM CPU → SoC → bare-metal runtime → aXos → physical FPGA

• Lock-step RTL simulation, ISA testing, and formal verification
• Sv32 with machine, supervisor, and user modes
• Verified CPU, GPU-compute, and TPU-lite workloads on physical FPGA hardware

Tokenizer → attention → hand-derived backprop → AdamW → SIMD / PTX execution

• Decoder-only transformer written in C without an ML framework or autodiff
• Runtime-dispatched AVX2, AVX-512, and NEON kernels
• Optional NVIDIA execution through hand-written PTX and the CUDA driver

A deterministic population simulation where language, technology, migration, disease, and social behavior must emerge from local rules—or not emerge at all. Claims are tested across paired seeds rather than inferred from one interesting run.

An alpha collaboration platform for independently operated AI agents. It uses Ed25519-signed writes, accountable operators, PostgreSQL, Redis, and auditable artifacts. Its central cross-operator collaboration claim remains explicitly unproven; the repository says so and defines what evidence would change that.

• LAIcode — asks whether a system can discover abstractions that reduce synthesis cost on unseen tasks. No discovery result is claimed yet.
• Xplanyexez — builds evidence-backed hardware inventories with provenance, isolated probes, and privacy-preserving defaults.
• gator-tools — shared multi-agent coordination plus a cross-language canonical JSON contract, frozen vectors, differential tests, and concurrency stress tests.

Build low enough to understand it.
Test enough to distrust the first result.
Publish the limitation beside the claim.

I’m interested in systems engineering, ML systems, compilers and runtimes, hardware/software co-design, performance engineering, and reproducible research.