$ whoami → kevin · ai engineer · ~/.portfolio

I build agents that run themselves, and the infrastructure underneath them.

A working portfolio of systems I've built — agent orchestration, evaluator-controlled loops, home infrastructure, on-device ML. Each one is written out so you can judge the engineering, not just the screenshot.

// six project examples below — read the ones that look interesting

## selected_work

Agent Engineering: For Apps, For Agents, For You

Universal Agent — agent_orchestration · the flagship
// agent_orchestration · the flagship

Universal Agent

Complete "Do Anything" Self-Driving Operator System. 140+ services on one runtime VPS; the desktop is only the cockpit. The system that builds and operates everything else.

The desktop is the cockpit. The VPS is the brain.
→ read case study
QLoop — evaluation · non-deterministic work
// evaluation · non-deterministic work

QLoop

Eval Harness for non-deterministic processes to facilitate autonomous long-life agent runs. The controller decides; the evaluator only scores — that split is what makes unattended loops trustworthy.

A long-running agent is only as trustworthy as the thing that decides whether its work is done.
→ read case study
SonosKD — home_infrastructure · the template
// home_infrastructure · the template

SonosKD

Home Server plugin for Whole-Home Audio Controller. Built as the reusable template every future home service copies — with a narrated explainer video (generated by ClearSpring Studio).

Not a Sonos remote. A home-server framework that happens to play music first.
→ read case study
Tome — on_device_ml · the alignment problem
// on_device_ml · the alignment problem

Tome

Narrated E-Reader with Uni-fetch sourcing. EPUB → Kokoro TTS → word-by-word highlight. The hard part isn't the reader — it's aligning the highlight to the voice.

A narrated reader is easy to fake and hard to do right. The gap is one word-timing alignment.
→ read case study
ClearSpring Deep Research — research · map-driven retrieval
// research · map-driven retrieval

ClearSpring Deep Research

Enhanced research system with Universal Artifact Generation. Builds a knowledge map first, then drives NotebookLM source discovery against it — report, audio, and video from one run.

A research agent without a frame produces a bibliography. With a map, a literature.
→ read case study
ClearSpring Studio — video_generation · dogfooded
// video_generation · dogfooded

ClearSpring Studio

Interview-Driven Video Generation for All Purposes. Idea to narrated, brand-matched MP4 in one session. Made the video embedded in the SonosKD case study — and it's the service ClearSpring sells.

The beat sheet is the swap point. Substitute a new one and re-run.
→ read case study
How these hang together. The same tailnet serves Universal Agent's surface, the render box, and SonosKD — it's a portfolio-wide operational default, not a per-project feature. QLoop grades the work Universal Agent produces; Deep Research and HyperFrames are skills that graduate into the same plugin library. Each project is stronger because the others exist.
## about

I build agents that operate themselves — the orchestration, the evaluation that makes them trustworthy, and the infrastructure they run on. These six systems are the proof.

Every case study names its real structural decisions, states its limitations honestly, and points at what it actually proves. If you want to talk about any of it — building, operating, or extending this kind of work — the door is open.

// or directlykevin@clearspringcg.comSubmissions forward to my inbox — I read every one.