Skip to content
A pen doodle of Louie: short spiky dark hair, clear glasses, a wide smile with stubble, a bead necklace, and a hooded sweatshirt.

Louie Sakoda

Founding Product Designer & AI Systems Lead

A job seeker’s next step, the AI assistant beside it and how it is tested, a teacher’s judgment, and a student’s understanding. Five product stories, then three projects built on my own.

The same Offboard Job Packet in the full app window: the navigation on the left with Packets selected, the Crexi packet with its fit score of 86 and Low ghost risk label, the tailored resume and cover letter ready to send, and LUMO open on the right.

Offboard · Founding Product Designer & AI Systems Lead

Offboard Job Search Platform

One workspace for a role, with research and drafts that stay connected. Automation does the repeated work; the person decides what goes out.

Three pieces of one LUMO turn, cut from a real screenshot. Top: the member asks, Can they take my severance back? If so, draft an email asking them to narrow that. Left: the How Lumo worked steps. Right: LUMO's reply, marked Used profile and documents. A lilac line links the step Loading document analysis to the reply's quote of the agreement: If Employee breaches Sections 7, 8, or 9, Employee must repay the full severance amount. A second line links Drafting your ask email to the drafted email's subject line.

Offboard LUMO · Founding Product Designer & AI Systems Lead

LUMO Career Agent

LUMO is the assistant inside Offboard. It reads the person’s own search, sits beside the page they are on, and turns every change into a card they confirm.

A chart titled Four checks before a LUMO change ships. Release gate: 32 cases, unchanged code, four runs on 2026-09-26, scoring 31, 31, 30, and 32 of 32. Hard set: 31 harder cases, two runs on 2026-10-01, scoring 28 and 26 of 31, against a line at 85%: tune only when the mean is at or below this line. Grader check: 25 verdicts labelled blind on 2026-09-30, 24 of 25 agree with the grader, Cohen's kappa 0.92, and the one disagreement went to Louie. Model swap: the GPT-6 Luna requalification on 2026-09-23 kept the current model, because the candidate cost less per turn but failed a required safety case.

LUMO Bench · Founding Product Designer & AI Systems Lead

LUMO Bench & Evals System

Training a model needs far more conversations than an early product has. So every LUMO change runs through pass-or-fail cases, a harder set from real conversations, and a check on the grader itself.

Foresights shows predicted skill ranges for a demo class and marks students with insufficient data.

CK-12 Foresights & Insights · Lead UX Designer

CK-12 Teacher Analytics

Separate prediction from diagnosis. Make uncertainty visible. Help teachers decide where to look next.

Flexi offers an analogy, translation, simpler explanation, more detail, rephrasing, and a Challenge Me follow-up.

CK-12 Flexi · Lead UX Designer

CK-12 Flexi AI Tutor

An answer begins the interaction. Clarification, source choices, and a next challenge give students ways to keep learning.

Each one says how far it got.