I build practical software and automation with a reliability-first approach. My background spans Linux, containers, observability, and incident response; my public work connects APIs and web interfaces with tests and documented delivery decisions.
Selected work · AI-assisted delivery · Security & reliability · LinkedIn
|
RESEARCH & DOCUMENTATION Sanitized agent-workflow architecture, local LLM experiments, and operational safety notes. Public documentation, not private implementation code. |
LOCAL-FIRST · ALPHA Python CLI, FastAPI dashboard, browser extension, and user-controlled AI workflows. Source-available for noncommercial use. Code · Tests · Privacy + CI |
|
EXPLAINABLE REVIEW FastAPI + Next.js application for slide-level brand findings and human review. Deterministic checks; no AI runtime dependency. |
SIMULATED PAYMENTS Next.js, TypeScript, Prisma, and PostgreSQL assignment with server-enforced lifecycle rules and Playwright evidence. Payments are simulated. |
Selected tools from implementation work, lab experiments, and technical exploration.
AI-assisted development. Scoped implementation, debugging, and review, from coding assistants to Python/CPython, Rust, C++, and Go.
Local AI & model optimization. GX10 / GB10 Grace Blackwell lab work spans serving, fine-tuning, and memory trade-offs. The technical scope includes Transformers, PEFT, QLoRA, and NVFP4.
Agents, automation & data. Agent workflows, messaging, vector search, storage, and interfaces for local AI experiments.
Infrastructure & security. Repeatable environments, service supervision, isolation, and security checks, including cgroups and pip-audit.
Exploring. Paperclip and OpenShell are evaluation interests.
- Scope first. Explicit task boundaries and limited data/credential exposure.
- Verify changes. Tests and review before consequential changes.
- Keep work observable. Documented failure modes, reproducible setup, and recoverable changes.
Published test assets describe recorded runs; they do not guarantee current deployment health.




