Igor Rogovik

Igor Rogovik

Product designer × AI · Toronto · Open to new roles

Currently: designing and prototyping AI tools end to end, from research and flows in Figma to working builds. Below are case studies with the process, the trade-offs and the numbers.

Igor Rogovik
← Home
Portrait
About

I like turning messy problems into tools people actually use.

A short, human paragraph: how you got into design, what draws you to AI, and the kind of problems you want to work on next. Keep it to three or four sentences.

Focus

Product design
AI interaction
Prototyping

Tools

Figma · Webflow
Claude · Claude Code
Python · HTML/CSS

Looking for

Product and UX roles
AI design roles
Toronto or remote

Igor Rogovik
← Home
Get in touch

Let's build something thoughtful together.

hello@yourname.com
Igor — Portfolio
← All work
Case study 01 · AI product · Automation · 2026

job(less)-bot: an AI job-search assistant that never hits send without me

Every application meant rewriting the same resume and answering the same questions in new words. I designed and built a pipeline that finds roles, tailors resumes and cover letters from my real experience only, and fills in forms, then waits for my approval.

Role
Solo: product design, UX and build
Timeline
2026 · in daily use
Stack
Python · Claude API · Playwright · Streamlit
Built with
Claude Code as coding partner
Thumbnail · dashboard review queue
01 / Context

The setup

User problem

Each posting wants a tailored resume, a cover letter and a form full of familiar questions. Generic AI tools are fast, but they invent experience.

Target audience

Me first: a job seeker targeting electrical technician roles in the GTA. More broadly, anyone who wants AI to do the busywork without losing control.

Goals

Cut the time per application, keep every claim truthful, and keep the final decision to submit in human hands.

Constraints

About $20 a month for AI, a free job API capped at 200 requests a month, and no scraping or bot-detection workarounds.

02 / Process

How it evolved

V1
Find jobs
→
Tailor resume
Proved AI could pick the right bullets, and showed it needed guardrails.
V2
Find
→
Free filters
→
Score · Haiku
→
Tailor
→
Honesty check
Cheap steps first. Code checks every claim against the master resume.
V3
Discover
→
Score
→
Tailor
→
Cover letter
→
Fill form
→
You approve
Forms are filled, but nothing submits without a click.
#
Obstacle
Resolution
01
AI can invent experience.
A plain-code validator: every bullet and figure must exist in my master resume.
02
Letters sounded like AI wrote them.
A style guide from my own writing, plus checks for phrases like “passionate”.
03
Some sites block automated browsers.
No evasion. Those jobs open in Firefox with a one-click Autofill script.
03 / Impact

The resolution

$0.0001
to score how well one job fits
$0.02
to tailor a resume to one posting
0
applications sent without my approval
AI cost per job, by step · to scale
Score fit (Haiku)
$0.0001
Outreach email (Sonnet)
$0.01
Tailor resume (Sonnet)
$0.02
To fill in
Add your real results: applications sent, interview rate, and time per application before and after.
Igor — Portfolio
← All work
Case study 02 · Desktop UX · Firmware · Open source

ChameleonUltra Unleashed: from command line to one-click RFID tool

The Chameleon Ultra is a capable RFID research device, but it's driven from a terminal and was missing common access-card types. I extended the open-source firmware and designed a desktop app that runs with a double-click.

Role
Solo: UX, GUI design, firmware
Timeline
2026
Stack
C on nRF52840 · Python · PySide6
Type
Open-source fork · GPL-3.0
Thumbnail · Unleashed app + device
01 / Context

The setup

User problem

Stock firmware couldn't read AWID or Paradox cards, a bug broke HID Prox detection, and everything ran from the command line.

Target audience

Security researchers, hobbyists and people testing access systems they're authorised to test.

Goals

Add the missing card types, fix HID Prox, and put every common task one or two clicks away.

Constraints

Fixed hardware with a limited HF chip, one exclusive serial port, and Windows users with no Python installed.

02 / Process

How it evolved

V0
Firmware v2.2.0
→
CLI commands
HID Prox misread. No AWID or Paradox. Terminal only.
V1
HID Prox fix
→
+ AWID
→
+ Paradox
→
Tkinter prototype
New card types verified on hardware. A quick GUI proves the flows.
V2
Firmware
→
Threaded device layer
→
Qt app
→
One-file .exe
Rebuilt for responsiveness. Double-click to run.
#
Obstacle
Resolution
01
HID Prox cards read with the wrong length.
Traced to an inverted guard-bit check. A one-line fix brought detection back.
02
The UI froze while the device was busy.
All serial I/O moved to one worker thread that talks to the UI through Qt signals.
03
Indala needs a demodulator the chip lacks.
Prototyped, then cut. The app says plainly what this hardware can't do.
03 / Impact

The resolution

+3
card types added or fixed
11
card types in one app
1 file
to install, no Python needed
To fill in
Add real-world signals: release downloads, GitHub stars, and a short screen recording of scan → emulate.