Newxel
Newxel
Newxel yra IT personalo ir inžinerinių paslaugų teikėja, padedanti įmonėms kurti ir valdyti dedikuotas programinės įrangos kūrimo komandas bei veikiančius tyrimų ir plėtros centrus. Jos paslaugos apima IT darbuotojų paiešką, personalo pajėgumų papildymą, nuotolinius plėtros centrus ir teisinę, finansinę bei operacinę pagalbą plečiant inžinerinę veiklą daugiau nei 10 šalių. Newxel bendradarbiauja su interneto ir mobiliųjų programų kūrimo, dirbtinio intelekto ir mašininio mokymosi, didžiųjų duomenų, žaidimų kūrimo, kokybės užtikrinimo, įterptinės programinės įrangos, kibernetinio saugumo ir blokų grandinės sričių specialistais, taip remdama ilgalaikes verslo partnerystes ir paskirstytas technines komandas.

Machine Learning Engineer, Computer Vision (Hybrid, Poland)

Own production computer vision systems for WSC Sports’ live video platform, from model evaluation through deployment. Improve GPU inference, benchmarks, and validation across sports video workloads.

Aprašymas

  • Lead development of video-understanding capabilities, including detection, tracking, recognition, and scene analysis.
  • Improve GPU inference throughput and cost with batching, precision tuning, and compilation.
  • Track production quality and latency, investigate failures, and resolve regressions.
  • Create and maintain labeled sports benchmarks organized by sport, league, and content type.
  • Set and monitor model metrics such as mAP, F1, precision, recall, IoU, latency, and GPU cost.
  • Compare models systematically and reproduce published results using company data.
  • Prepare decision memos covering recommendations, trade-offs, and licensing.
  • Maintain regression testing and shadow or A/B pipelines to validate model and configuration updates before release.
  • Follow research papers, model releases, and leaderboards to identify candidates for evaluation.

Reikalavimai

  • At least 3–5 years delivering production machine learning systems, including 2+ years in computer vision; video experience is preferred.
  • Strong Python and PyTorch skills.
  • Practical experience profiling and optimizing GPU inference with TensorRT and ONNX Runtime.
  • Hands-on experience in at least two areas: object detection and segmentation; OCR and scene-text recognition; or video action recognition and temporal localization.
  • Experience creating test sets, calculating mAP, F1, and temporal IoU, and conducting rigorous model comparisons.
  • Ability to critically assess research papers and model cards and reproduce published findings.
  • Production deployment experience with Docker, queue-based services, and cloud GPU infrastructure.
  • Experience processing video at scale, including FFmpeg, decoding pipelines, and frame sampling.
  • Strong written communication and the ability to document technical trade-offs and make clear recommendations.
  • Preferred: Azure Service Bus and Blob Storage, message-driven architectures, vision-language models, multimodal understanding, domain-specific vision fine-tuning, self-supervised learning, synthetic data generation, CVAT, Label Studio, shadow deployments, A/B testing, or commercial model licensing.

Privalumai

  • Own the complete machine learning lifecycle, from reproducing research and creating benchmarks to optimizing live GPU workloads.
  • Deliver production improvements at scale, reducing infrastructure costs and latency through model optimization.
  • Work in an evidence-led engineering environment that emphasizes A/B tests, shadow deployments, and statistical validation.

Susiję darbai

EY

AI Architect at EY (Hybrid, Katowice)

EY

Shape governed, scalable AI solutions for EY clients, from strategy and architecture through implementation. Lead initiatives across generative AI, LLMs, MLOps, cloud, and data architecture.

Atidaryti