Build the robot that shows up tomorrow morning.

01Why this role exists

EVE-01 picks ripe strawberries autonomously. It works because a tightly-tuned perception stack feeds two robotic arms through a ROS 2 motion planner — every pass through a row hardens the data and improves the next harvest. We are looking for the engineer who owns that loop end-to-end: from photons hitting the camera, to bounding boxes on a Jetson, to a clean cut at the calyx.

The team is small enough that what you ship today runs against real plants in our in-house lab the same afternoon, and big enough that you will see your metrics move on a per-row basis.

02What you'll do

• Own the perception pipeline. Detection (Ultralytics YOLOv8/v11), ripeness classification, occlusion handling, depth fusion against OAK 4D stereo. You set the targets and you ship.
• Train, deploy, repeat. Capture greenhouse footage → label in Roboflow → version datasets → train → benchmark → TensorRT-export to Jetson → A/B in the field. Move F1 and mean precision up without ballooning compute.
• Write production ROS 2. Nav2 stack on the differential-drive base, MoveIt 2 trajectories for the dual arms, custom nodes for sensor fusion and fault detection. ROS 2 Humble today; Jazzy when it makes sense.
• Bridge sim and real. Isaac Sim scenes matching our in-house lab, synthetic data for hard cases (low-light, dense canopy), sim-to-real validation.
• Bring up new hardware. Moving from Jetson Orin to Jetson Thor, you'll port the inference stack and re-tune for the new compute envelope.
• Work hands-on with the robots. The greenhouse lab is on-site — some hours you are in code, some you are in front of a plant debugging a fault with a tablet on a workbench.

03Required

• 3+ years building production ROS 2 (or hard ROS 1 → ROS 2 migration experience). Nav2 and MoveIt 2 specifically — not theoretical, you have shipped with them.
• Trained and deployed at least one CV model that mattered. YOLO, RT-DETR, Mask R-CNN — whichever. What matters is you have taken something from labeled data to a constrained runtime.
• Comfortable in C++ and Python. ROS 2 nodes lean C++; the data/training stack is Python. You do not have to love both equally, but you are not blocked by either.
• Linux, CUDA, Docker, Git — the unglamorous foundation.
• Field-engineering temperament. Greenhouses are humid, dusty, and full of bees. Robots break. You debug them on-site without complaining.

04Strong plus (any subset)

• Ultralytics ecosystem (ultralytics, YOLOv8/v11) and the Roboflow workflow — datasets, augmentation, model assistant.
• DepthAI / OAK 4D / Luxonis — you know what depthai-python does and what an IMUPacket is.
• Isaac Sim / Isaac Lab for synthetic data generation and policy training.
• TensorRT / ONNX optimization — int8 quantization, layer fusion, hitting latency budgets on Jetson Orin / Thor.
• NVIDIA Jetson bring-up: JetPack, nvarguscamerasrc, DLA offload, thermal headroom on a 40W module at 35°C ambient.
• MLOps for edge: Weights & Biases / MLflow / DVC, OTA model deployment, rollback when a new model regresses pick rate.
• Surgical-robotics or cable-driven actuation background. We don't expect it; if you have it, we want to hear from you.
• ELRS / mavlink / radio control — the manual-override side of the robot.

05What we offer

Time

• 32-hour work week, Monday through Thursday, full pay. We have found that engineers building hard physical systems do better work on a four-day cadence.
• Long weekends, every weekend. Three full days off, every week, no exceptions.

Money

• Competitive salary set against the senior Tbilisi engineering market, plus meaningful equity. Range disclosed in the first call.
• Premium health insurance for you and your immediate family — private clinics, no copay headaches.

Workspace & equipment

• Nice office in central Tbilisi. Designed for focus — quiet desks, real meeting rooms, a workshop floor, and the in-house greenhouse lab one door away.
• Top-spec MacBook Pro M4 Pro as your workstation, day one. Plus whatever monitors, peripherals, and field gear you need.
• Unlimited coffee, tea, cookies, and fresh fruit in the kitchen. The strawberries are, occasionally, our own.

Tools

• Claude Max subscription for every engineer, plus Cursor / GitHub Copilot / whatever AI tools speed you up. We invest in tooling, not in slowing you down.
• NVIDIA developer accounts, full Jetson hardware on your desk, all the cloud GPU credits training needs.
• Conference & training budget — ROSCon, NVIDIA GTC, CVPR, whatever pulls you forward.

If you're relocating

• Work permit & relocation support for non-Georgian hires. A work permit is required to legally work in Georgia — we sponsor and walk you through the full application, including any residence-permit steps.
• Relocation stipend covering flights, first-month rent, and a temporary apartment while you find a place.

06Living in Tbilisi

Tbilisi is a real city. Walking-friendly, dense, dramatically cheaper than Western Europe or California, surrounded by mountains, vineyards, and an 8,000-year wine-making tradition. The tech scene is small but growing fast — you will run into the same people at meetups twice a month. English is widely spoken in the technology community.

The team eats together at a long table on Thursday evenings before our long weekends start. Black Sea coast is three hours west. The mountains are two hours north.

07How to apply

Email hr@neurapickrobotics.com with three things:

1. Your CV or LinkedIn — short is fine.
2. One artifact you are proud of: a GitHub repo, a paper, a Loom of a robot doing something hard, a Roboflow project, anything.
3. One sentence on which greenhouse-harvesting problem you would want to own first.

We respond within 5 business days. Field trials are continuous, so we are hiring on a rolling basis — no application deadline.