AI products should earn trust before anyone depends on them. That’s the part I work on.

• Tested complete performance review workflows, from employee self-evaluation through feedback collection and evaluator review to report generation.
• GenAI and voice AI testing: validated AI-driven conversations and voice-based survey interviews with AI agents, including transcript generation, transcription behaviour, follow-up questions, and the generated reports and workflow outcomes.
• 360-degree feedback and review cycles: tested respondent management, survey launch and monitoring, review-cycle statuses and the transitions between stages.
• Complex defect finding: identified and documented functional, UI, workflow and data-integrity defects, including incorrect survey and status updates, missing actions, respondent visibility problems, inconsistent workflow states and report-generation issues.
• API and full-stack investigation: validated APIs and investigated issues across frontend and backend workflows to find where each one started.
• Release validation: ran smoke and regression tests across QA and development environments, validated deployments, and performed production sanity checks on critical functionality.
• Frontend development: build frontend features in TypeScript and Node.js, and test what I ship.
• Collaboration: worked with developers to reproduce defects, verify fixes and improve application stability.

• TypeScript
• Node.js
• Voice AI testing
• API testing
• End-to-end testing
• Regression testing

5,000+users20+countries~3 minsmall reports, down from 15–2050+component bugs found30%less manual testing

• End-to-end, regression, smoke and release testing across UI, database and production for the List Builder, Project Chat, Reports, Artifacts and Admin modules.
• Faster reports through an Azure Service Bus fix: small reports took 15–20 minutes and large ones about 30. While testing report generation at the cloud level, I found an Azure Service Bus autoscaling issue. After the fix, small reports finish in about 3 minutes and large ones in 5–7.
• Lower AI cost per report: large reports that used the Excel query MCP tool were costing $40–50 each in model usage. Monitoring OpenAI, Anthropic, Gemini and Serper usage during testing helped trace the root cause, and after the fix the cost per report dropped drastically.
• Load testing: found that report generation and artifact generation were the bottlenecks under load, and the team improved both.
• Component library testing: found 50+ rendering bugs in the shared library that every screen’s components are rendered from, including how components come out when downloaded as PowerPoint (PPT).
• Regression automation: built Playwright (Python) regression automation with pytest and Allure, using AI-assisted development to write scripts faster, cutting manual testing effort by 30%.
• Release ownership: handled 5–6 QA and production deployments, from release R3.6 to R4.4, covering branch cuts, environment variable updates, post-deployment validation and production smoke tests.
• Tenant setup: created 5 tenants, including 1 production and 3 QA tenants, with configuration and database-level validation for each.
• Production incident: traced a live production issue through application logs and Datadog, restored service within about 2 hours by restarting the container as a temporary fix, and documented the root cause for a permanent fix.

• Playwright (Python)
• pytest
• Allure
• JMeter
• Azure
• Datadog
• GitHub Actions

11K+users13+countries~35 minfull regression run40%more smoke coverage

• Manual and automation testing for releases used by 11K+ people in 13+ countries.
• Regression automation framework: regression for the platform was slow and manual, so I automated the high-priority document analysis, canvas and meeting-notes flows in Python and Playwright, then refactored them to a Page Object Model with class-scoped pytest fixtures and a shared browser context. A full run now takes about 35 minutes.
• Smoke suite with dynamic test data: replaced hardcoded projects with a fresh project created on every run, and added negative and edge-case scenarios, giving about 40% more coverage and a clean environment each time.
• Reporting and CI: integrated Allure reports organised by epic, feature and story, with screenshots and step logs, and set up a headless UI automation pipeline in Azure DevOps, so runs are repeatable and results are readable by non-engineers.
• Flaky-test debugging: traced failures to environment changes, timing and fragile locators such as index-based selectors and screenshot timeouts, and replaced them with stable locators in a central locator module, so previously failing tests now run reliably.
• Load testing for file uploads: built JMeter scenarios for 100-file, 200-file and multi-user uploads with automatic token refresh, analysed bottlenecks in Celery’s background processing, and documented the setup, giving developers concrete performance limits to work on.
• Real-time (WebSocket) testing: tested multi-tab, multi-user, slow-network and large-batch scenarios while inspecting console and WebSocket traffic, and found critical status-sync bugs, each documented with video, logs and exact reproduction steps.
• UI/UX revamp validation: tested every major module end to end and re-tested fixes alongside developers as they deployed, logging 60+ issues and shortening fix turnaround.
• Test design for complex features: wrote test plans and positive, negative and edge-case scenarios from requirements, Figma designs and documentation for role-based access control, data retention, multi-language support, assistant creation and AI-generated documents.
• UAT triage and release readiness: reproduced UAT-reported issues across environments, classified each as a bug, enhancement, environment-specific, duplicate or not-a-bug, wrote UAT guides and reviewed release notes, which cut back-and-forth between QA, developers and UAT.
• API validation and exploratory testing: used Postman to confirm fixes directly against the API before the UI was ready, explored an AI deep-research feature, and found a duplicate-submission bug affecting many modules, catching issues before production.

• Python
• Playwright
• pytest
• Page Object Model
• Allure
• Azure DevOps
• JMeter
• Celery
• WebSockets
• Postman
• Jira

• Built Playwright (Python) smoke and regression suites with Allure reporting for an AI web application in the private equity domain.
• Manual testing: test case design, execution, defect reporting, and reproducing issues with developers and product.