Nir Ben Zvi

Email Website LinkedIn GitHub X Phone

Professional Experience

ML/AI Consultant and Fractional Head of Research (2021-present)

  • Freelance applied ML scientist and research lead spanning vision, video, language and multimodal systems.
  • Owning the full machine learning lifecycle, from problem definition, data and modeling through evaluation, production code and deployment.
  • Working with companies ranging from early-stage startups to large public corporations, from hands-on research to fractional research leadership.
  • Training and deploying state-of-the-art open-source models, including on multi-node, multi-GPU systems.
  • Production work includes:
    • Computer Vision: diffusion/flow models, VLMs and video models, segmentation, detection and classification.
    • Language/LLMs: complex agentic systems, retrieval/RAG, harness design, LLM evaluation suites and benchmarks, and on-prem/remote model serving.
  • Partial list of clients:
    Keep (Acq. by Elasticsearch), Mixtiles, Simply, Deci.AI (Acq. by NVIDIA), Lemonade ($LMND), CommonGround (Acq. by Apple), DeepChecks, Oddity ($ODD), Papaya-Global, Honeybook, Clover Security, and many more.

Computer Vision Research Manager, Trigo (2018-2021)

  • Founding engineer and the first computer vision hire.
  • Built Trigo’s first deep learning models from scratch, running in real-time in live supermarkets.
  • Led development of object detection and real-time tracking models optimized for deployment.
  • Scaled machine learning R&D from garage phase (<10 employees) to 100+ employees, hiring and growing top-tier ML researchers and engineers.
  • Built annotation pipelines, ML infrastructure, and model evaluation tools to support large-scale CV development.

Computer Vision Research Scientist, Amazon Lab126 (2015-2018)

  • Researched, trained and deployed computer vision models for Amazon’s consumer devices, including the Echo Look and the Echo Show.
  • Led research efforts for the group’s first deep-learning-based object detection models used in production.
  • Worked end-to-end on ML research, from whiteboard concepts to production-grade implementations in Amazon products.
  • Central technical reference point for computer vision research and infrastructure within the org, evaluating emerging research, methods and tooling.
  • Contributed to Amazon’s (discontinued) MXNet deep learning framework.

Computer Vision Algorithm Engineer, Donde Search (Acquired by Shopify) (2015)

  • Coded Donde’s first visual search engine using Caffe.

Imagineer, Disney Research (2013-2014)

  • Research associate under supervision of Prof. Arik Shamir.
  • Researched style-transfer algorithms based on real artist data (pre-deep learning).
  • Developed image contour tracking models for artistic style transfer.

Education

  • M.Sc., Computer Science, The Hebrew University of Jerusalem (2013 - 2015)
  • B.Sc., Computer Science, The Hebrew University of Jerusalem (2010 - 2013)
  • Minor in Visual Communications, Bezalel Academy of Art and Design (2010-2014)

Publications

Alexander Lorbert, Nir Ben-Zvi, Arridhana Ciptadi, Eduard Oks, Ambrish Tyagi
Toward Better Reconstruction of Style Images with GANs
KDD2017 Workshop on Fashion AI

Nir Ben-Zvi, Jose Bento Ayres Pereira, Moshe Mahler, Jessica Hodgins, Arik Shamir
Line-Drawing Video Stylization
Eurographics 2016

Other

Teaching

  • Intro to Object Oriented Programming, The Hebrew University of Jerusalem (2014)
  • MDLI Deep Learning CS231 Community Course, Google Campus (2017-2020)

Military Service

Full military service in the elite unit of the IDF artillery corps, “Moran” (2006-2009)

  • Squad-commanders course graduate.
  • Reserves duty: In charge of the unit’s assessments of new recruits (“gibush”).

Code/Tools

  • 10+ years of experience writing Python research and production code on top of POSIX systems.
  • Data processing; postgresql (inc. pgvector), bigquery, snowflake as well as current vector-dbs (currently in love with Qdrant).
  • Thorough knowledge of deep learning hardware, compute/data pipelines and model throughput optimization (ONNX, TRT, etc.).
  • Production experience on AWS and GCP for ML workloads.
  • Full proficiency with common 3D, video and image editing tools (Blender, Photoshop, Lightroom, Illustrator).
  • Hardware geek; spec’d (and built!) enterprise-grade GPU rackmounts for ML training/inference.

Inter/Personal

  • Full English fluency; Native Hebrew speaker.
  • Ex-professional Photographer.
  • A generally nice guy.