AI Engineer - US Remote

Maneva•Atlanta, GA

8h•$125,000•Remote

About The Position

Maneva is hiring an AI Engineer to define and build vision models for its Vision-to-Action AI platform, which uses video from existing cameras to support real-time decisions and actions in manufacturing environments. The models will run on production lines, where accuracy, latency, and reliability are critical. The role covers technical strategy and architecture across computer vision detection, vision-language models (VLMs), and vision-language-action (VLA) workloads. The engineer will train and adapt models using proprietary data, develop data and labeling systems, create rigorous evaluation methods, and deploy optimized models on cloud GPUs and edge hardware. The position also leads cross-team technical work, mentors engineers, and evaluates emerging research for potential product use. The role may support projects subject to U.S. export controls (ITAR); access to some projects may require U.S.-person status or export authorization.

Requirements

At least 5 years of experience building and shipping ML systems, with a focus on computer vision.
BS, MS, or PhD in computer science, electrical engineering, or a related field, or equivalent experience.
Deep hands-on experience with object detection and segmentation, including YOLO-family or DETR-style models, covering training, tuning, and production deployment.
Hands-on experience with VLMs, including fine-tuning, prompting, grounding, and evaluation.
Experience with VLAs or embodied and robot-learning models, or a clear record in adjacent multimodal work.
Strong PyTorch skills and fluency with distributed training, mixed precision, and GPU profiling.
Track record of taking models from research to production, including optimization, inference serving, monitoring, and iteration on real-world failures.
Evidence of staff-level impact, including setting direction across teams, making build-versus-adopt decisions, and raising the bar for others.
Clear written and verbal communication with engineers, researchers, and product leaders.

Nice To Haves

Publications or open-source contributions in vision or multimodal learning.
Experience with video understanding, 3D perception, or multi-sensor fusion.
Experience with simulation, sim-to-real transfer, reinforcement learning, or imitation learning.
Edge and embedded deployment experience, including TensorRT, CUDA, or quantization-aware training.
Experience with LLM tooling such as vLLM or RLHF-style post-training for multimodal models.

Responsibilities

Define technical strategy and architecture for vision models across detection, VLM, and VLA workloads.
Train, fine-tune, and adapt models on proprietary data, including distillation, quantization, and LoRA-style adaptation.
Build data engines, including labeling pipelines, foundation-model-assisted auto-labeling, active learning, and synthetic or simulated data.
Design evaluation benchmarks, regression suites, failure-mode analysis, and online metrics tied to product outcomes.
Deploy models to production within latency, memory, and cost constraints on cloud GPUs and edge hardware.
Lead cross-team technical projects, write design documents, review code and models, and mentor senior and mid-level engineers.
Track developments in research, assess what is worth adopting, and turn selected approaches into prototypes and products.
Optimize, serve, monitor, and iterate on production models based on real-world failures.