Full-Stack Applied AI Engineer

JAY SNEDDEN

Techy in, revenue out.

Built, broken, and verified by hand.Explore the work

01 / ABOUT

About

I'm Jay Snedden. I've been programming since I was eight, starting old-school: FreeBSD, systems work, whatever language the problem needed.

Over my IT consulting career I worked with roughly 3,000 business clients. I'm a tech consultant and trainer by trade: I teach technical concepts from basic to complex, and I help teams turn that understanding into results.

For the last few years I've gone deep on AI and agentic workflows, building most nights in a home lab: several servers at home plus two VPS boxes in Boston. Local models are my specialty. I test the way I was taught to engineer: relentlessly, and against reality.

My guiding philosophy, which I teach and try to live by:

“Being inquisitive, not assumptive, leads us to true knowledge and understanding.”

And my rule for everything on this site: I don't trust paper, I verify architecturally.

I study from first principles. Levenshtein's edit distance and Zadeh's fuzzy logic aren't citations in my work, they're what my data-cleaning pipelines are built on.

What I like most is talking with real people about real problems. If you're working on something and want a straight answer, that's my favorite kind of conversation.

02 / PROJECTS

Projects

Short outlines of things I've built. Full write-ups where they exist, notes where they're still in progress.

Local Model Consistency Harness

Most model evaluations tell you what a model said once. I wanted to know what it says a thousand times. This harness runs roughly 1,000-iteration sampling runs to measure output consistency and variance between model variants, tests for bit-flip anomalies, and evaluates abliterated models. If you're deploying a local model into anything that matters, variance is the number nobody gives you. This measures it.

In plain English: before you trust an AI model with real work, I can tell you how consistent its answers actually are.

Home-built NOC/SOC

A from-scratch network and security operations setup for my own infrastructure. A node agent runs on every Linux machine and reports to a central UI: thermal tracking per unit of compute, kernel and stack versions across the fleet, and full system inventory. It was built for centralized agentic control of the whole lab, not just dashboards.

In plain English: I built my own mission-control dashboard for every computer I own, so one system watches them all.

Agent Harnesses

Custom agent setups I've built and run daily: OpenCode, OpenClaw, Hermes, and a Pi-based harness of my own design. Access is gated by port-knocking tied to the router. These are working tools, not demos, and the notes here cover what survived contact with real use.

In plain English: I build AI assistants that do real multi-step work, and I've tested the frameworks hard enough to know where each one breaks.

Automated AI Podcast Pipeline

Two daily shows generated end to end by a pipeline I built: research, scripting, multi-voice synthesis, QA (including loop and repetition detection on the audio), a 10 percent speedup master, and RSS publishing. The morning brief runs solo and rapid. The deep dive brings three hosts to the two or three biggest stories, each with a verdict: real, overhyped, or watch this space.

Both shows are rebroadcast on my own feed. Listen

In plain English: I built a machine that researches, writes, voices, and publishes two daily shows by itself, every morning.

More on the way

Write-ups in progress. Check back, or subscribe to the podcast feed for the audio version of what I'm working on.

03 / PAPERS & WRITING

Papers / Writing

I write to figure out what I think. Papers, essays, and long-form notes land here.

No papers published yet. This page will fill in as write-ups are finished.

04 / PODCAST

Podcast

Two shows, every morning, generated by a pipeline I built and broadcast from my own feed. No third-party platform required.

I built this to brief myself. Press play to hear today's edition.

All episodes and subscribe options

One subscription carries both formats.

01 / THE BRIEF

Jay's Daily AI Brief.

About eight minutes. One voice, rapid rundown: the headline, plus one line on why it matters. I listen while I brush my teeth, and by the time I'm done I know how much today's AI news actually matters.

Also in your podcast app

02 / THE DEEP DIVE

Jay's AI Deep Dive.

About thirty minutes. I built this one to keep up with AI myself: the two or three biggest stories, argued over by three hosts with three different perspectives. Every story ends with a verdict: real, overhyped, or watch this space. Good high-level listening for the drive to work. I use it every day. Thought I would share.

Also in your podcast app

05 / CONSULTING

Consulting

A note on what this is, and what it isn't.

I'm an independent engineer, not a firm. I take on a small number of private consulting engagements, working directly with teams and individuals. There is no agency here: no account managers, no sales process, no retainers. You talk to me, and I do the work.

I like conversations with real people about real problems. If you're trying to figure out whether AI can actually help with something, that's my favorite kind of conversation, and I'm happy to have it before anything is for sale.

What I help with:

Local and small-compute AI.: Getting real work out of models on hardware you actually own, including consistency testing before you deploy.
Agent systems.: Designing, testing, and hardening agentic workflows, and telling you honestly where they break.
Technical translation.: Turning deep technical concepts into language your team, your leadership, or your customers can act on.
Verification and evaluation.: If someone sold you an AI system, I can tell you what it's actually doing.

How it works: short, focused engagements with a defined question and a written result. I keep availability limited and I say no when the fit isn't there.

If that's useful to you, write to jay@jaysserver.ai with a paragraph on what you're trying to do.

06 / CONTACT

Contact

The fastest way to reach me is email:

jay@jaysserver.ai

I read everything myself.

For consulting inquiries, include a paragraph on what you're trying to do.