Kevin BartolomeAI Engineer · Forward Deployed Engineer · Lead Software QA Engineer · Software Developer

US Citizen · Remote or relocation
kevin.rl.bartolome@gmail.com
LinkedIn · kxvin1.github.io ↓ Download PDF

Engineer who builds AI agents that investigate and fix failing end-to-end (E2E) tests for customers of QA Wolf, an AI testing platform, and leads its largest team, 13 engineers covering about 200 customers, while still investigating hands-on. Former QA lead for Harvey, a $15.5B AI legal platform.

Work Experience

QA Wolf

Lead Software QA Engineer

10/2025 to Present

Lead the company's largest team, its only full-time investigation team: 13 hand-picked top performers (large teams were 6 or 7) investigating 5,000+ E2E test failures a day, and growing, for about 200 customers, including DoorDash, Harvey, Figma, Grafana, Metronome and Braintrust
Cleared 300 to 500 failing E2E tests a day on Harvey, in about four hours, with AI agents I built: they read logs, traces and live DOM to a root cause with evidence, which lets me lead this team and still investigate hands-on all day
Cut SLA breaches about 40% and company-wide investigation time about 28% with a fixed daily on-call routine for customer messages and written SLA rules applied the same way for every customer, giving the company's own customer teams back roughly 300 to 375 hours a week
Built the new team's processes from scratch at a fast-moving startup: had it running in under a working day, set its standards and how it uses the company's AI, and trained every engineer on it, raising the share of E2E test suites that AI closes on its own from about a third to about half
Led QA for Harvey, a $15.5B AI legal platform, before this team: owned E2E test strategy for its Playwright and TypeScript suite of 1,600+ E2E tests and a strong relationship with its engineers and CSMs, the record that got me picked to lead this team

QA Wolf

Software QA Engineer

07/2025 to 10/2025

Promoted to Lead Software QA Engineer within three months of hire
Made AI output testable: 65 E2E tests that check non-deterministic LLM behavior (agentic document workflows, streaming completions, citation integrity) even though answers change every run
Extended E2E coverage past the browser to Microsoft Word (web and PC), iOS/Appium, email, Twilio SMS/MFA and Google OAuth via a stealth browser, reaching flows browser-only E2E tests miss

Zeta Global

Software Developer

02/2023 to 07/2025

Retired a standing manual reporting job with internal React/Next.js dashboards on real-time PostgreSQL filtering, adopted across teams as their working view into the data
Replaced manual data gathering with one validated pipeline: a Python collection service over REST APIs and SQL that improved the accuracy of the data teams reported on, tested and deployed through GitHub Actions and Docker

Projects

Mission Control

Human-in-the-Loop AI Automation Personal project, 2026 to Present

kxvin1.github.io/mission-control

Built an AI dashboard that investigates hundreds of failures at once, 10x as efficiently as by hand: agents group the failures, work the groups in parallel via QA Wolf's CLI and MCP, and verify each fix before I apply it; a bug report is one keystroke, not 20-30 minutes
Created 50 fully vetted E2E tests in one day, against a company-wide expectation of 10 a week, with creation and review agents I built: every written step must map to an assertion seen passing, and a second agent challenges every review verdict
Ran 20 client investigations at once, about 800 failures in all, with scale built in: a lead agent session per client, worker sessions that each own a piece of work end to end, and all state in plain-text records, so no work is lost when a session ends
Made the agents safe to leave running on investigation and creation: evidence gates stop them from self-certifying ("ready" needs a clean local run plus a run on QA Wolf's infrastructure, same code, run id attached), and guardrails refuse destructive platform actions at the tool boundary
Cut context cost from 3-8k tokens per raw page snapshot to ~1k per filtered report, and live-DOM inspection from 62s to 17s, with every request, reply and model call in an inbox with a derived status and a per-day cost ledger

Skills

AI & Agents: Multi-agent orchestration, LLM evals, MCP, tool calling, guardrails, human-in-the-loop, context engineering, prompt engineering, AI observability, structured output, Claude Code
Languages: TypeScript, Python, JavaScript, SQL, Bash
Testing: Playwright, Page Object Model, Appium, E2E test automation and strategy, root-cause analysis, failure triage
Web & Backend: React, Next.js, Node.js, REST APIs, PostgreSQL
Infrastructure: Docker, GitHub Actions, CI/CD

Education

App Academy1,000-hour full stack web development, 07/2022 ·CSU NorthridgeBachelor of Science in Psychology, 12/2019