Volodya Dolzhenko

Founder and sole engineer across product, infrastructure, distribution, and QA. The Windows desktop app combines input tracking, gameplay recording, replay/VOD review, and post-game analytics, localized into five languages.

• Built a voice AI analyst (an OpenAI tool-calling agent) over MicroLab’s game-telemetry database, and the LLM eval pipeline around it: DeepEval golden-dataset evals with LLM-as-judge metrics for correctness, refusal, and faithfulness, deterministic tests asserting tool choice and arguments, and prompt-injection cases, all running in GitHub Actions with per-question scoreboards on every run.
• Traced every agent run with Langfuse, attaching judge scores to traces for pass-rate trending over time. Judge calibration tests grade each judge against hand-written answers with known verdicts; the evals caught three real agent defects before release, each fixed and locked in with regression tests.
• Designed the product’s test suite and CI pipeline (~65 tests): unit tests, golden-dataset regression, contract tests against real Cloudflare Workers storage (D1, R2), and Playwright end-to-end tests driving the packaged Electron app, running on every pull request.
• Built on an Electron client with a serverless backend on Cloudflare Workers (D1, R2), plus Stripe billing, Riot Games API integration, Sentry crash reporting, and signed auto-updating releases.
• Cultivated a close community of over 300 users: gathering feedback, working through their suggestions, chasing down reported bugs, and shaping the roadmap around what players actually asked for.
• Grew an organic international user base across 13 Riot regions, with 2,500+ games uploaded and analyzed to date.

Promoted from manual QA to automation lead within four months, then grew into senior test engineering over three roles.

• Led QA for an LLM document-intelligence pipeline that OCR’d legal documents and extracted structured financial data, benchmarking GPT and Claude models against a 95% accuracy pass-rate bar and dropping any that fell below it.
• Built automated TypeScript validation of LLM outputs, enforcing JSON-schema conformance and probing hallucination and failure modes with edge-case suites and a golden-dataset regression baseline.
• Wrote API and GraphQL tests validating a wide range of endpoint outputs, plus visual and frontend tests covering 40+ public-facing partner pages.
• Tested event-driven architectures end to end: injecting mock events into Kinesis streams and validating the results downstream in the data warehouse to catch silent data corruption.
• Standardized the org on Cypress and Playwright, defined QA KPIs, and cut regression cycle time by 50% with risk-based test planning. Ran load and stress testing with k6 and Artillery at 1.5 million requests per hour across five parallel runners.

I wanted to see whether I could get past Wild Rift’s anti-cheat. Every software option I looked at left a fingerprint on the device, so I went fully external: a video capture card isolates the phone’s video output, and nothing touches the game itself.

From there it became a latency problem. In competitive play, reaction speed is everything, and my research showed that routing anything over a network added too much delay, so I needed to emulate a touch with the lowest latency possible for it to be any good.

A second computer runs a computer-vision script that watches the isolated feed and fires an action once a condition is met (in my case, a monitored number dropping below a set threshold). The trick was the tap itself. Capacitive touchscreens register a finger because it changes the charge at that point on the glass, so I stuck a small pad of copper tape to the screen and ran a thin wire from it back to a pin on the microcontroller. When the script decides to act, the controller drives that pin to shift the charge on the copper pad, and the phone reads it as a real fingertip touch. No motors, no robotic finger, just an electrical nudge at the right spot.

It worked, though the setup was fairly complicated, and I never ended up using it in a real match. I was satisfied all the same: it was my first hardware project, and I learned a lot, including soldering, which turned out to be really fun.