Computer Vision jobs

Position Title: AI / Computer Vision Engineer

Role Type: Full-Time

Location: On-site

Role Overview

We are looking for a hands-on AI / Computer Vision Engineer to design, train, fine-tune, and deploy state-of-the-art vision and vision-language models. In this role, you will work directly with foundational computer vision architectures, optimize real-time inference pipelines, and handle edge/GPU deployment. Experience with human-in-the-loop workflows and active learning pipelines is a major plus.

Key Responsibilities

• Model Training & Fine-Tuning: Custom-train and fine-tune foundational vision and vision-language models (YOLO, DINO, SAM, VLMs) for specific domain applications.
• Pipeline Optimization & Real-Time Inference: Optimize vision pipelines for high-throughput, low-latency execution and real-time streaming.
• Model Deployment: Convert, quantify, and deploy models using runtime frameworks (ONNX Runtime, TensorRT, Triton) on GPU and edge environments.
• Data & Active Learning: Integrate human-in-the-loop (HITL) processes and active learning pipelines to continuously improve dataset quality and model performance over time.
• Integration: Collaborate with engineering teams to integrate computer vision modules into backend services and production architectures.

Required Technical Skills & Qualifications

Core Architectures: Hands-on experience with:

• YOLO (v8, v9, v10, or v11 for object detection and instance segmentation)
• DINO / DINOv2 (Self-supervised vision transformers)
• SAM / SAM 2 (Segment Anything Model for interactive/promptable segmentation)
• Vision-Language Models (VLMs) (e.g., CLIP, Qwen-VL, LLaVA, Florence-2)
• ML Lifecycle: Proven track record in model training, fine-tuning, loss function optimization, and handling dataset edge cases.

Deployment & Inference:

• Proficient in model export and optimization via ONNX, TensorRT, or OpenVINO.
• Deep experience with CUDA/GPU acceleration and real-time execution limits.

Frameworks & Tooling:

• Strong skills in PyTorch, OpenCV, and modern annotation/dataset platforms.

Nice-to-Have Experience

• Human-in-the-Loop (HITL): Designing feedback loops where human annotators review uncertain predictions to iteratively train models.
• Active Learning: Implementing automated data selection methods (e.g., uncertainty sampling, core-set selection) to maximize annotation efficiency.
• Edge AI Deployment: Deploying on target devices (e.g., NVIDIA Jetson, embedded Linux).

Work Location: In person