Automation architecture · Python engineering · Agent reliability
I’m an automation consultant with over a decade of experience across web, mobile, APIs, and enterprise workflows. I design test architectures, diagnose difficult failures, and build tools that make engineering decisions easier to verify.
My current work brings that discipline to AI agents: capture what happened, check the outcome, and make failures understandable.
Portfolio & experience · The Automation Architect’s Playbook · LinkedIn
- Automation architecture: framework design, browser and API testing, mobile automation, and CI delivery.
- Failure diagnosis: UI stability signals, visual comparisons, test reports, and reproducible evidence.
- Agent reliability: controlled LLM tests, execution records, explicit agent contracts, and outcome verification.
Reduce timing-related Selenium failures with configurable UI stability checks and diagnostics. Application-specific assertions still decide whether the outcome is correct.
pip install waitless · Technical walkthrough
Fixture-scoped mocks for supported LLM SDK calls, including typed responses, streaming scenarios, and repeatable failures. Active fixtures intercept supported calls; installing the plugin alone does not block network access.
pip install "pytest-mockllm[openai]" · Testing boundaries
HTML reports for pytest and unittest with test-phase results, step details, and screenshot evidence to help investigate failures.
pip install pytest-glow-report · Project walkthrough
Save and restore current-origin cookies and web storage, with optional encryption and explicit restore validation. Session reuse complements dedicated login tests.
pip install selenium-teleport · Session reuse guide
A general-purpose agent platform for turning goals into inspectable outcomes.
Starlight coordinates domain agents working with code, data, APIs, browsers, devices, or language models. It bounds mission execution and returns reports with agent identity, attempts, results, and evidence. Agents own their domain logic and outcome checks.
The current 5.x alpha includes a Node.js 22+ runtime, local CLI, and authenticated remote-agent protocol. A browser or model provider is not required. The earlier browser-specific implementation is legacy.
Repository · Demo & documentation · Start with the architecture
- Agent Blackbox: a local-first flight recorder and CI diagnostic gate for AI-agent runs.
- Project Vandal: runtime UI mutation testing for Playwright to check whether tests catch regressions.
- Visual Guard: screenshot baselines, visual comparisons, region masking, and reviewable differences.
- SB Stealth Wrapper: a SeleniumBase UC Mode reliability wrapper for authorized testing, with bounded recovery and explicit failures.
- Visual Sonar: computer-vision automation for WVD and Citrix environments without DOM access.
- The Automation Architect’s Playbook: a free guide to framework design, browser internals, failure analysis, and AI-assisted testing.
- My writing on Waitless was featured in PyCoder’s Weekly #714.
- My background includes published research in spatial clustering, which informs how I approach algorithms and automation design.
Working through an unreliable test suite, an automation architecture decision, or an AI-agent workflow that needs better evidence?