Ethan Olsen
Ethan OlsenFlagstaff, AZ · NAU Electrical Engineering

Electrical engineer · MS candidate, Northern Arizona University Software engineer · Founder, Laplacient

I design circuits, model power systems, and teach machines about silicon. I ship full-stack products and agentic AI workflows.

I've done full-wave Ansys HFSS modeling of signal waveguides in advanced semiconductor packages, and transmission-line models of power converters. I like problems where the physics and the math both have to hold up. FastAPI backends, React frontends, Postgres job queues and Claude-powered agents, shipped for my own company and for consulting clients. I also spend my days finding where frontier LLMs break.

This site has two sides. Pick one:

Buck converter · switch node● RUN
~/projects

            
  • 3.78GPA, BS + MS in Electrical Engineering at NAU
  • FounderLaplacient: AI products and consulting
  • AI TrainerSurge AI, finding frontier-LLM failure modes since 2023
  • Ansys HFSSSignal integrity for advanced packaging, ASU research 2026

Selected work

EE 599 · Graduate power electronics · Nov 2026

Distributed buck converter with a lossy transmission line

I reproduced Röbenack & Gerbet (2024) in Python, derived the duty-to-current transfer functions from the telegrapher's equations, and extended them with an Lin ∥ Rp input element. Damping with Rp ≈ Z₀ cut the worst-case input ripple at the operating point from 16.3 A to 0.22 A.

  • Python
  • NumPy
  • Scilab / Xcos
  • LTspice
  • Multisim

ASU SWAP Hub · Semiconductor Solutions Challenge · Apr 2026

IR-drop prediction with a fully convolutional U-Net

A PyTorch U-Net that predicts power-delivery-network voltage drop from current, PDN density and effective-distance maps. It was pretrained on synthetic layouts and fine-tuned on 10 real circuits. On unseen real designs it reached RMSE 0.54 mV and hotspot precision 0.73.

  • PyTorch
  • U-Net
  • Dense regression
  • Data pipelines

MCM / ICM 2026 · Problem A · Team 2629908

"Why does my phone keep dying?" Smartphone battery model

A continuous-time, physics-based additive model of display, CPU, radio and streaming power, calibrated against real iPhone XR drain logs. The headline finding: radios matter. A Wi-Fi-only day lasts ~34 h, and the same day on 5G only lasts ~17 h.

  • Mathematical modeling
  • Python
  • Calibration

Power systems · Mar–Apr 2026

Grid-connected multi-string PV inverter

A two-stage MATLAB/Simulink model: dual DC/DC boost converters with hill-climbing MPPT, a PID-controlled DC link, and AC grid injection.

  • MATLAB
  • Simulink
  • MPPT
  • PID control

Laplacient · Founder · 2025–present

Laplacient Social: an AI social media manager

A Claude agent researches angles, writes per-platform captions, and generates images and video. It schedules through PostPeer, but only after a human approves. Background agent turns survive restarts, and idempotent scheduling means retries never double-post. Stripe credits meter the cost.

  • FastAPI
  • React + TS
  • Claude tool use
  • Neon Postgres
  • Cloudflare Workers + R2
  • fal.ai
  • Stripe

Senior capstone 2026–27 · Project Manager · Sponsor: U.S. Army Yuma Proving Ground

AME: Automated Metadata Extraction

A catalog for test-data file shares. It scans directories, extracts timestamps, sensor type, format and units with per-field provenance and confidence, and serves search and export through a web UI and REST API. Everything ships as one offline Docker image.

I'm the project manager of a six-person team (four CS, two EE). I lead sprint planning and our Jira board, own the GitHub repo and merge workflow, review every PR, and built the single-container deployment.

  • Python
  • FastAPI
  • PostgreSQL 17
  • React + Vite
  • Docker
  • Jira / Agile

Laplacient · Consulting

Learn Prompt Engineering

A consulting practice that helps small businesses adopt AI: scoping sessions, workflow automation and custom builds. Client work includes an SEO automation platform for CadenceSEO and an invoicing agent pipeline for Ecorial, each saving 5+ hours a week.

  • Astro
  • TypeScript
  • Agentic workflows
  • Client discovery

Lab bench

Every model on this page is real code from my projects, ported to run in your browser. Change the inputs and check the results yourself.

Signal integrity · Ansys HFSS research

Designing a via-fenced stripline for a package substrate

This is the structure I modeled most in HFSS: a trace buried in dielectric between two ground planes, with rows of stitching vias tying the planes together. The trace width sets impedance. The via pitch and fence spacing set how high in frequency the line stays clean before energy leaks into parasitic modes between the planes.

Power electronics · transmission lines

Why a 6 m cable breaks a buck converter, and how one resistor fixes it

Replacing the converter's inductor with a real coax line adds resonances in the PWM band. At 8.49 MHz the line rings and the input ripple reaches amps. A small inductor in series at the input, damped by a resistor near the line's Z₀, absorbs those reflections.

Mathematical modeling · MCM 2026

What actually drains your phone?

P(t) = Pdisplay + PCPU + Pradio + Pstream + P₀, calibrated per app from two days of measured drain. Change one setting for the whole day and see how long the battery lasts.

Machine learning · chip power integrity

Predicting IR-drop hotspots on circuits the model never saw

Aggregate on the 10 hidden real circuits: RMSE 0.54 mV · MAE 0.45 mV · NRMSE 0.118 · hotspot precision 0.729 · recall 0.573 · F1 0.510 (hotspot = top 10% of true drop).

AI training · Surge AI

Spot the failure

My day job is finding the sentence where a frontier model goes wrong, in EE, software and math, and writing down why. Try it: each answer below has exactly one failure.

Product engineering · Laplacient Social

From a form to a scheduled post, with a human in the loop

Each step shows the plan and post state machines and the parts of the stack involved.

Systems design · Capstone

One container, offline, no message broker

The client network has no internet access, so AME ships as one image: Postgres doubles as the job queue, and a plain bash entrypoint under tini supervises everything.

Consulting · Learn Prompt Engineering

Turning a weekly chore into a review step

I work out what a team repeats every week, automate it, and leave a person in charge of the decisions. Flip each card to compare the workflow before and after.

Experience

  1. Jun – Jul 2026 · Tempe, AZ

    Advanced Semiconductor Packaging Research Intern Arizona State University

    • Built and ran Ansys HFSS full-wave models of package signal waveguides (coplanar waveguide and via-fenced stripline between two ground planes), interpreting S-parameters, Smith charts and impedance matching.
    • Studied EMIB and CoWoS for high-frequency die-to-die links.
    • Did cleanroom work (lithography, etching, CVD) at ASU NanoFab and MacroTechnology Works, and wire-bond testing on the Sigma Bond Tester.
  2. Mar – Jul 2026 · Gilbert, AZ

    Software Engineering Intern CadenceSEO

    • Led a team of three engineers building internal tools that automate page optimization and SEO analytics.
    • Owned the SDLC (PR reviews, task allocation) and production deploys on Vercel, Supabase, Railway, Render and Heroku.
    • Managed API key usage, spend limits and rotation.
  3. Sep 2025 – present · Remote

    Founder Laplacient

    • Built Laplacient Social, which automates social post generation, scheduling and analytics.
    • Runs Learn Prompt Engineering, turning client goals into designs and implementations (CadenceSEO, Ecorial).
  4. Aug 2023 – present · Remote

    AI Trainer Surge AI

    • Research failure modes of frontier LLMs in electrical engineering, software engineering and mathematics.
    • Write adversarial conversations and failure analyses covering correctness, behavior and writing quality.
    • Build reproducible evaluation environments with Docker, Bash and Harbor CLI, and write reference solutions in high-frequency signal analysis, full-stack engineering and analytic number theory.
  5. 2021 – 2024

    Before engineering

    Shift manager at McDonald's (running crews, scheduling, delegation), structural ironworker and SMAW welder at Arc Steel, and drop-in math tutor at Chandler-Gilbert Community College.

Let's build something

I'm looking for roles and projects where hardware and software meet: power and signal integrity, ML for EDA, or AI products that need someone who reads the datasheet.