Skip to content
View vladbondarenko97's full-sized avatar
🎯
Focusing
🎯
Focusing
6 followers · 8 following

Achievements

Achievement: Pair ExtraordinaireAchievement: Pull Sharkx2Achievement: YOLO

Achievements

Achievement: Pair ExtraordinaireAchievement: Pull Sharkx2Achievement: YOLO

Highlights

  • Pro

Block or report vladbondarenko97

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
vladbondarenko97/README.md

Typing SVG

I test software that can't afford to fail: surgical devices, mortgage platforms, healthcare systems and, lately, AI agents.

website linkedin email

How I work

  • 🧪 An AI agent is a system under test. I measure what it did, not what it says it did. My browser agent reported 25/25 when 23 were true; I caught that, fixed it, and made the fix measurable.
  • 🛡️ Safety lives in code, not in prompts. Allowlists, budget caps and policy hooks that the model cannot talk its way past, enforced in CI.
  • 📏 Evidence decides. A cheaper model only gets a job if it passes the same quality gate as the expensive one.
  • 🏥 Regulated by default. Six years in FDA-regulated medical devices, HIPAA healthcare and mortgage lending. Every requirement traces to a test.

➡️ How I engineer: the design decisions, the bugs I root-caused, my test strategy and how I keep coding agents honest. Every claim links to the code.

Featured work

🛞 Take the Wheel

Type a task; Claude drives your real, logged-in Chrome tab, with a human in the loop for logins and payments.

  • $0.33 → $0.034 per task, 29 → 4 turns, accuracy 23 → 25 / 25
  • Found that the agent's allowlist wasn't the real boundary (it exposed GitHub credentials and local files) and replaced it with a deterministic PreToolUse guard, tested across Haiku, Sonnet and Opus
  • Runs on a Claude plan or AWS Bedrock with a model allowlist

Claude Code Chrome MV3 CDP Node.js Bedrock

📋 ReqTrace

An auditable AI pipeline for regulated engineering: SRS → cited findings → trace matrix → draft verification protocol.

  • 5 / 5 planted defects, 0 false positives, gated in CI
  • Moving one step to a local 9B model cut model cost 55% with no quality loss; 3 of 4 cheaper routings failed the gate and were rejected
  • Flags unapproved edits to controlled documents, by two independent mechanisms

Python Pydantic LLM evals local models · live dashboard

📈 Market Terminal

A self-hosted market desk that never shows a number it can't stand behind.

  • 426 offline tests, plus CI that runs the clock 400 days ahead
  • Every raw response stored and replayable to the byte; a failed source shows as missing, never zero
  • Coding agents (Claude Code, Codex, Cursor) work under one rulebook, and CI fails the build when one breaks a rule

Python SQLite Flask local LLM

🌍 vlad.yt

My portfolio: a scroll-driven WebGL descent from the Sun to Earth's inner core.

  • Ships with its own Playwright suite on desktop and phone, axe accessibility checks and a no-WebGL fallback test
  • CI refuses to deploy unless the committed bundle equals a fresh build, then deploys over a restricted SSH key

three.js GLSL Playwright GitHub Actions

Open source

🎮 AnyPS5: macOS host support · merged #7 and #8

  • Traced a PS5 title failing on macOS to two missing libkernel imports, then to a crash on its first C++ exception when the unwinder hit an unimplemented call
  • Implemented kqueue/kevent and sceKernelGetModuleInfoFromAddr for macOS, with guest regression tests and unwind fixtures; full 712-test suite green before maintainer review

C++ Python CMake macOS systems debugging

Tech stack

Test automation

Playwright Cypress Selenium Cucumber Postman Karate Ranorex axe

AI & agents

Claude Claude Code Bedrock Local LLMs Evals Guardrails

Languages

Python JavaScript TypeScript Java C# SQL Kotlin Swift

CI/CD & infra

GitHub Actions Jenkins Docker AWS Ansible Node.js Flask

Quality management

Jira Zephyr RequisitePro TDD

Regulated domains

FDA HIPAA Mortgage WCAG

Experience

Company and role Years
👁️ Bausch + Lomb, Software Verification Engineer, R&D 2026 → now
🏦 U.S. Bank, QA Business Technology Analyst 2019 – 2022
🩺 Centene · Express Scripts · Answers.com, QA Automation Engineer 2017 – 2019
🌡️ Emerson, Mobile Developer (Sensi smart thermostat) 2016 – 2017

Full story, from the Sun to the inner core: vlad.yt

GitHub stats

Pinned Loading

  1. take-the-wheel take-the-wheel Public

    Type a task, Claude drives your real logged-in Chrome tab. Cents per task, human-in-the-loop handoffs, and a model-independent security guard on every agent action, enforced in CI.

    JavaScript 3

  2. market-terminal market-terminal Public

    Self-hosted market data pipeline and web terminal: options, macro risk, forecasting models. Python, SQLite, Flask.

    Python

  3. vlad-yt vlad-yt Public

    Scroll-driven WebGL portfolio: from the Sun to Earth's inner core. Playwright e2e tests + CI.

    JavaScript