The Role:
Researches, architectures, and constructs production AI vision systems, leveraging state-of-the-art detection models and stream processing pipelines to extract attributes and infer intelligence from image and video streams in real time.
Responsibilities:
Model Architecture & Development: Design, train, and fine-tune state-of-the-art vision models (e.g., YOLO variants, DETR, Transformers) for object detection, classification, and attribute extraction.
Stream Processing & Pipeline Engineering: Build low-latency ingestion and processing pipelines for high-throughput video and image streams (e.g., RTSP, WebRTC, frame decoding).
Plate Recognition (LPR/ANPR), Optical Character Recognition (OCR), and multi-object tracking (MOT).
Inference Optimization: Optimize model performance for efficient execution on cloud infrastructure (Kubernetes/GCP) and edge devices using TensorRT, ONNX, or OpenVINO.
Dataset & Active Learning: Curate, annotate, and augment visual datasets, establishing automated model retraining loops and evaluation frameworks.
Requirements:




3+ years of hands‑on experience designing and deploying computer vision and deep learning models in production environments.
Deep expertise with modern object detection architectures (YOLO, DETR, Faster R‑CNN, Vision Transformers).
Strong programming skills in Python and deep learning frameworks (PyTorch, OpenCV, TensorFlow).
Proven experience working with video stream ingestion, frame processing, and low‑latency inference.
Demonstrated experience with License Plate Recognition (LPR), OCR, or fine‑grained attribute inference systems.
Familiarity with model quantization, ONNX export, and TensorRT compilation for hardware acceleration.
YOLO / DETR | Computer Vision | PyTorch / OpenCV | Video Stream Processing | LPR / OCR |
Disclaimer – Must read: Commitment & Focus
This role requires full‑time dedication, with clear priority given to Darwoft projects during the established working hours. It is not compatible with othe

📌 On-Site It Systems (Perú)
🏢 Amazon
📍 Perú