RF hardware engineer/Space & defense/Denver, CO
RF hardware engineer.
AI-native builder ofeverything else.
I lead RF architecture and front-end design for space and defense hardware: link budgets, L- to Ku-band front ends, phased arrays and flight test. Outside the lab, AI agents are my engineering team. With them I ship RF test benches, a mains-switching PCB, software products and the infrastructure they run on.
Key specifications
| RF hardware for space and defense | 7+yrs |
| Engineering hours saved by one tool I built | 1,500hrs |
| Builds shipped with AI agents, below | 11 |
| Parts on a mains PCB, brief to routed in 3 days | 288 |
| REF | CR1 | U1 | FL1 | U2 | MX1 | U3 | CASCADE |
|---|---|---|---|---|---|---|---|
| G, dB | −0.4 | +18.0 | −1.8 | +15.0 | −7.0 | +20.0 | +43.8 |
| NF, dB | 0.4 | 0.9 | 1.8 | 2.5 | 7.5 | 3.0 | 1.42 |
| NF to here | 0.40 | 1.30 | 1.33 | 1.39 | 1.41 | 1.42 |
The day job: RF systems for space and defense.
A1
System architecture
Link budgets, architecture trades and RF architecture for high-stakes proposals, and presenting the design to customers as the technical authority.
A2
Front-end design
L- through Ku-band front ends, from HFSS, Microwave Office and ADS simulation through procurement and flight test.
A3
Phased arrays
Digital beamforming boards, a 1:16 distribution network at 35 dB gain and 40 A, and bare-metal T/R switch firmware.
A4
Test and process
SCPI test automation, standardized design reviews and "test like you fly" protocols that cut non-conformances.
Before engineering, nine years in operations and supply-chain planning. It's why I weigh trades in schedule and dollars as well as dB.
Full resume →Software, hardware and everything in between.
Each build was designed and shipped with AI agents, mostly Claude Code, under the same discipline I use on flight hardware. Every page says what the agents did and how the work was checked.
02.1
RF & test
Instrument control and analysis tools for the RF lab, built for how RF engineers actually work.

RF Workbench
Browser RF tools on this site. Draw a signal chain and get a live cascade of gain, noise figure, IP3, P1dB and sensitivity; load Touchstone files and compare parts on rectangular or Smith axes. Nothing is uploaded.
- 150
- unit tests on the RF math
- 2 + 1
- parallel agents + me
- 0
- bytes uploaded

PA Characterization Bench
A browser-driven RF power-amplifier test bench. It runs a vector signal generator, a signal analyzer and a 4-channel supply over raw SCPI, and turns a morning of manual measurements into one button per test.
- 71
- commits in 12 days
- 107
- independent math checks
- 7 → 0
- SCPI errors per sweep after bring-up

S-Parameter Viewer
A VNA-style Touchstone viewer for comparing parts. Overlay up to ten .sNp files, one-click return loss, insertion loss or isolation, snap-to-point markers, cross-grid trace math, and slide-ready export.
- 1
- day, first commit to deployed
- 10
- files overlaid at once
- N-port
- any port count
02.2
Hardware
Boards and mechanical parts, from mains-rated PCBs to parametric 3D-printed designs.

4-Zone Fermentation Controller
A mains-switching brewery controller designed as a real, manufacturable product. ESP32-S3, four triac heater zones and four relay pump zones, 288 parts on a 4-layer board. Taken from written brief to routed PCB in three days, with AI agents driving KiCad.
- 288
- components, 9 schematic sheets
- 43
- logged design decisions
- 0 / 0
- ERC errors / warnings

Parametric Tabletop Token System
A fully parametric OpenSCAD system of 3D-printable grid tokens, with 47 base tiles that drop flush into ring frames for Large, Huge and Gargantuan creatures. The build script exports and verifies 191 meshes, and a Blender scene built entirely in code renders the product shots.
- 191
- meshes built and verified
- 47
- tiles fit one print bed
- 4.0
- mm, flush assembly

Two-Door Guinea Pig Cottage
A parametric, 3D-printable hideout designed around animal behavior and printability. Five parts, no fasteners, no glue, no supports, and a build script that refuses to package a non-manifold mesh.
- 0
- supports, fasteners or glue
- 1.27
- kg PETG, five prints
- 0.35
- mm dovetail fit
02.3
Software
Products in real use: AI vision and voice, invoicing, e-commerce.

InVinetory
Point a phone at a wine label and it identifies the bottle, researches it once for every cellar, and files it in a shared household inventory. Describe dinner out loud and a voice sommelier picks from bottles you actually own.
- 13/14
- labels right on real photos
- 2.7 s
- median read, ~0.4¢ each
- 2 days
- for all 6 planned phases

TymTracker
A time-tracking and invoicing app I built when my invoicing tool's price went from about $100 to about $2,000 a year. It runs my own invoicing today, and is being rebuilt as a multi-tenant SaaS product.
- 20×
- price increase avoided
- 40
- tests, incl. auth against Postgres
- 22 → 8
- npm audit advisories

Tensor3D Storefront
A self-hosted e-commerce platform that replaced Shopify for my 3D-printer parts shop. It covers storefront, checkout, oversell protection, shipping labels, admin and analytics, and I migrated in four years of order history.
- 10
- days of build, 32 commits
- 983
- historical orders migrated
- 17 → 0
- audit findings open
02.4
Infrastructure
The self-hosted platform everything else runs on.
- SW1
10 GbE core
6 VLANs
- N1–N2
Storage nodes
NVMe + dual-path SAS
- N3
Media node
GPU passthrough
- N4–N5
App nodes
my apps + services
- U1
Edge
VPN + tunnel only
The Homelab
A five-node Proxmox cluster with 218 TB of ZFS, a segmented 10 GbE network, and around twenty services. It hosts every app on this site, and it's run as documented infrastructure with an AI agent that keeps the runbook current.
- 5
- cluster nodes, ~20 VMs
- 218 TB
- RAIDZ2, grown in place
- 6
- VLANs, IPS on each
- J1–J8
PoE cameras
RTSP, H.264 / H.265
- U1
go2rtc
WebRTC, no transcode
- U2
Backend
credentials, PTZ, passkeys
- U3
PWA
iPhone home screen
Vordr Camera Viewer
A self-hosted, view-only live viewer for eight PoE cameras that installs on an iPhone home screen. WebRTC with no transcoding, pinch-zoom and PTZ, and private cameras that unlock with a passkey. Nothing leaves the private network.
- 8
- cameras, live
- ~4%
- CPU for 5120×1920 + audio
- 0
- requests leave the network
Flight-hardware discipline, applied to AI agents.
The habits that keep flight hardware out of trouble (written requirements, gated reviews, traceable decisions, test like you fly) are what turn AI agents from fast into fast and right.
1
Brief
Requirements, constraints and what "done" means, written before any code or copper.
2
Phase gates
Architecture, parts, schematic, layout. Agents stop at every gate; I approve or send them back.
3
Decision log
Every non-obvious call recorded with the alternatives and the evidence, so no session starts blind.
4
Independent review
Separate agents audit the work: ERC and DRC, mains stress, security, datasheets, math from first principles.
RF Workbench
Draw a signal chain and get a live cascade of gain, noise figure, IP3, P1dB and sensitivity at every stage. Load Touchstone files and compare parts on rectangular or Smith axes. Runs entirely in the browser.
Ask about my work.
An assistant built on the Claude API that answers from this site's own pages, the resume and every case study, and links its sources. It's one more build: a small Cloudflare Worker with rate limits and a daily budget.
Ask · answers come only from this site
Try one of these, or ask your own: