Binh Vo —
Systems architect and hands-on builder focused on agentic AI infrastructure, local model tooling, retrieval systems, and the operating layer that turns ambiguous work into repeatable execution.
Working Notes
My work sits between architecture and implementation. I like the part of systems work where the requirements are still fuzzy, the data sources do not agree, the workflow is half manual, and the first useful version has to be designed while it is being built.
Lately that has meant private AI infrastructure, agentic workflows, retrieval-augmented systems, memory layers, source-authority rules, and automation that can survive contact with real operators. I am interested in AI systems less as demos and more as execution environments: tools with state, judgment boundaries, handoffs, observability, and a clear path from user intent to finished work.
Before that, I spent time in enterprise IT risk and creative operations, which shaped how I think about control design, access boundaries, documentation, media pipelines, and stakeholder translation. That background shows up here as a bias toward systems that are explainable, auditable, and usable by people who were not in the room when the architecture was drawn.
This site is where I make the work visible without turning it into a resume. Projects document systems I am assembling. Insights track what I am learning as the scaffolding changes. Setup notes judge tools by the failure modes that appear after the first clean install.
Approach
A public bench for private systems work.
The writing here is organized around the questions I keep running into while building AI systems: when should memory be hot state versus durable record, how should agents choose between tools, what belongs in a workflow engine versus application code, and how do you evaluate local models against actual work instead of marketing claims?
I avoid polished case-study theater. The useful parts are the design constraints, tradeoffs, experiments, and teardown notes that help another builder decide what to trust, what to replace, and what to leave boring.