Kyrylo Bielohryvtsev

Test Engineer, autonomous vehicles

I get parking lots and curbs ready for driverless pickup and drop-off.

I design the zones where driverless vehicles pick up, drop off, charge, and park, prove them in simulation, run the on-site tests, and coordinate the operators, engineers, and partners it takes to get each site to live service.

10+
depot lots built or supported
100+
destinations taken through readiness
~20
offshore team members I coordinate
Email me Download résumé
Animation: a pickup zone is drawn over three parking spaces, then a vehicle parks inside it and pulls back out onto the road. Pickup zone Simulation result Park Get-on-road

Illustrative lot. Yellow traces the park maneuver, white the pull-out back onto the road.

Experience at Zoox

  1. Jan 2025 – Mar 2026

    Advanced Testing Instructor & Operator

    • Validated 1,000+ autonomous test miles across general, pickup/drop-off, and highway missions.
    • Trained new software and vehicle operators on missions such as release validation, regression, stability, mapping, and geofence validation.
    • Four-month Los Angeles deployment to launch a new city, mapping much of the area.
    • Testing at multiple proving grounds; installed a weather sensor kit on test vehicles in Miami.

  2. 2026 – present

    Test Engineer

    One role with three areas of focus, working closely with a ~20-person offshore team, engineers, and leadership.

    • DepotsBuilt and validated zones for 10+ depot lots: charging, staging, storage, cleaning, repair, and pickup and drop-off.
    • DestinationsTook 100+ hotels, event venues, and airports across six cities through readiness.
    • Scheduling and coordinationScheduled testing and data collection for both, across operations, engineering, and partners.

Depot build-out

10+ depot lots across two cities, each with zones for charging, staging, storage, cleaning, repair, and pickup and drop-off. Most I took end to end through the three steps below; on others I provided on-site planner test support while another team owned the zone.

  1. lane segment

    1Draw the zone

    Draw each zone over the exact parking spaces and lane segments the vehicle will use, then land the zone files in the map through a reviewed PR.

  2. 1234 5678 Space Test vehicle Driverless ParkGet-on-road ParkGet-on-road pass triaged, re-run

    2Prove it in simulation

    Build park and get-on-road scenarios for every space, for both the test vehicle and the driverless vehicle. Triage failures from the logs, then merge the passing suite into the main code repository once CI is green.

  3. 3Test on site

    Write the test plan, request vehicles and operators from the operations team, and join the harder runs in person.

What I check on every lot

Select part of this illustrative lot to see what I'm responsible for there.

Parking spaces Pickup zone Drop-off zone Gate Ramp Drive aisle Sidewalk

Pickup zone

Where it's checked:

Geometry is illustrative. No real site, address, or partner is shown.

Beyond the geometry. I test the planner and motion behavior underneath each maneuver, such as reverse handling and path cost, and dig into logged vehicle messages to explain why a run behaved the way it did.

Planner support. On lots where another team owned the zone, I ran on-site planner testing for the harder driving evaluations.

Destination readiness at scale

Hotels, event venues, and airports each go through the same five phases before riders can use them. For the past six months I've driven 100+ destinations through that cycle: doing the zone, test, and review work myself and coordinating everyone around it.

Each dot is one destination.
Lobby Valet and taxis

Hotels

Entrance loops shared with valet, taxis, and deliveries. The zone has to sit where riders naturally wait.

Venue Closed on event days

Event venues

Big surges before and after events, and roads that close on event days. Zones need room for crowds.

Terminal Door ADoor BDoor CDoor D Accessible Zone aligned to Door C

Airports

The most complex to build: accessible loading zones, airline-specific doors, and curb rules worked through with the product team.

  1. Manual traversal

    An operator drives the site manually to capture a first log.

    Operations

  2. Build the zone

    I build many of these zones myself and merge them into the map. Complex sites get input from product and engineering.

    Built by me

  3. Test the zone

    Schedule autonomous test drives through the new zone.

    Operations

  4. Review the logs

    Check every pickup and drop-off, and that operators can dispatch the site from the app.

    Operations

  5. Sign off

    Hand off site-specific fixes, confirm the configuration, and release to live service.

    Engineering, leadership, partnerships

Why it takes coordination. A destination only goes live when partnerships, product, engineering, and leadership agree on the partner arrangement, how the zone is built, and how it's configured. I keep those threads moving across dozens of sites at once.

Data collection

Every depot and destination gets its own data-collection tickets, and each has different requirements. I write the test plans, brief operators, schedule the runs, and confirm the right data came back.

  • Zone accuracy

    A first check that a newly built zone matches the real curb and parking spaces.

  • Dispatch

    How the vehicle gets dispatched to and from the zone.

  • App hailing

    End-to-end rides requested through the mobile app.

  • After a zone change

    Fresh data whenever a zone is fixed or updated, to confirm the change works.

  • Gates and ramps

    Barrier arms, sliding gates, ramps, and other site features perception needs data on.

Collection tracker

Illustrative
SiteZone accuracyDispatchApp hailingAfter a zone changeGates and ramps
Depot A
Hotel B
Airport C
Venue D
Depot E
Hotel F

Collected ScheduledBlank: not needed for this site

Prototyping with AI

When the existing tools didn't show what I needed, I used AI coding assistants to build small tools on top of them: combining data sources, filling in missing features, and making results visible at a glance. They're prototypes for my own workflow, built on existing tools and data.

  • Claude Code
  • Plugins and skills
  • MCP servers
  • Cursor
  • Python
  • JavaScript
414243444546474849505152535455565758596061626364 pass fail not run

Results by parking space

Shows simulation pass or fail for every space on the lot map, so a failure at space 45 or 57 stands out right away instead of hiding in a results list.

L1 L2 L3

3D view for multi-level depots

Renders stacked floors, ramps, and overlapping road segments in 3D, which are hard to read in a flat map view.

Signed off Ready to review Signed off Testing Blocked Testing

Readiness tracker

Pulls ticket status, notes, and log links from Jira, Slack, Databricks, and spreadsheets into one board, so in-progress and blocked sites stand out without opening each ticket.

7.4 m 5.2 m Validate Export

Zone drawing experiment

Combined existing map and validation tools into one view, so parking spaces and lane segments stay visible while drawing a zone. An experiment alongside the official editor, not a replacement for it.

How I work

Most of the job is keeping the right people aligned so the technical work actually lands.

Test engineer Offshore team (~20)Test operatorsOperationsEngineeringProductPartnershipsLeadership I direct the work I coordinate with

Leadership

  • Coordinate a ~20-person offshore team on testing and review work
  • Brief and direct test operators on site
  • Coordinate operations, engineering, product, partnerships, and leadership
  • Schedule and triage testing across many sites at once

Test engineering

  • Test plans for on-site and data-collection runs
  • Zone design, simulation suites, and validation
  • Planner and motion testing: parking, get-on-road, reverse maneuvers
  • Log review, failure triage, and sign-off

Technical

  • Editing zone and config files in protobuf text format (pbtxt)
  • Git: task branches, creating and reviewing PRs, CI validation, merging
  • AI-assisted development with Claude Code (plugins, skills) and Cursor
  • Setting up MCP servers to connect AI assistants to Jira, Slack, and other team tools
  • Python and JavaScript for prototypes and automation
  • Jira, Agile, two-week sprints