Dmytro Hrybov

ml research engineer. eight years shipping deep learning into production, most of it computer vision. now in san francisco, building evals for physical-world ai.

now

  • since aug 2026

    founder resident at entrepreneur first, the bridge (s26), san francisco. benchmarking frontier vlms on physical-world tasks: real robot episodes compiled into replay-verified sim environments. earlier, fine-tuned minimax h3 (33b video diffusion) into a video+action robot policy on DROID.

before

  1. apr–aug 2026

    modelroom · co-founder & cto

    agents that plan and execute multi-shot video production, with a vlm harness scoring every shot for prompt adherence and cross-shot consistency. built the core, led four engineers.

  2. 2025–26

    google · software engineer, kaggle research

    reproduced and transparently hosted major llm benchmarks (mmlu, facts grounding) on kaggle benchmarks. built the agentic pipelines behind the data science agent in colab enterprise. led gpu region expansion, rolling out b200 / h200 / h100 capacity across gcp.

  3. 2023–25

    promaton · senior machine learning researcher

    3d mesh reconstruction and deformation, point-cloud segmentation and generation, pose estimation in the dental domain. ran several greenfield projects end to end: system design, data acquisition, research, deployment.

  4. 2020–23

    surgalign · machine learning engineer

    neural networks on 2d and 3d medical scans for surgical tasks: spinal cord instance segmentation, pedicle screw placement, 3d patient orientation. led a five-person team annotating ct and o-arm volumes. rebuilt the training pipeline for a 240% throughput increase.

  5. 2019–20

    tcl research europe · ai engineer

    image enhancement research for tvs and phones: hdrnet, gans, transformers, detection and segmentation. improved a human-body segmentation network by 4.09% mean iou. co-authored the ntire 2020 challenge on video quality mapping (cvpr workshops).

also

  • 2022–23

    real-time generative installations for science and art events. audio-to-video stylegan3 and a custom-optimized image-to-image sdxl turbo, both running live above 30 fps.

  • 2016–20

    bsc computer science (machine learning), polish-japanese academy of information technology, warsaw.

  • i speak english, ukrainian, polish and russian fluently.