Subtain Malik

S.B Malik

Senior AI/ML Engineer

Work experience
  1. 2024 – now Sigrow Senior Software Engineer, AI/ML · Netherlands · HortiEngine

    Owns the ML serving and data platform: 20 vision models, 8 million predictions a day, p95 27.2 s → 1.25 s. 2 million messages a day from 5K+ IoT devices.

  2. 2023 – 2025 AI/ML Consultant For Rapidev Group & Relymer

    OwlSense social-media intelligence on a multi-agent harness with local LLMs. 23× faster LLM inference; one serving layer for 100K+ requests a day, $50K a year saved.

  3. 2021 – 2024 DARVIS Machine Learning Engineer · United States, remote · OmniRoom

    Real-time video analytics across 100+ cameras, models running in parallel into one decision layer. +50% FPS from custom DeepStream plugins.

  4. 2020 – 2021 CENTAIC (PAF) Machine Learning Engineer · Islamabad · AIRAF

    AI detection on 450-megapixel satellite images, 95% accurate on objects under 10 pixels, with 80% less analysis time. Mission-critical work under strict security limits.

Other achievements
  1. 2024 PyCon Talk Python and the Multiverse of Cameras · PyCon Pakistan

    A tale of synchronization and streams, on stage at PyCon Pakistan 2024. The talk is on YouTube.

  2. 2023 Research Paper Reimagining UI Design using Deep Learning · arXiv · first author

    Challenges and opportunities of deep learning for app interface design. ICT Endowment Research & Development Funds award.

  3. 2022 OpenSource Contribution Merged upstream · TensorRTX · PytorchX

    ArcFace face recognition on TensorRT 8 in TensorRTX (Aug 2022), plus MLP examples in TensorRTX and PytorchX (Jan 2022).

  4. 2018 – 2021 Gold Medal MS Computational Science & Engineering · NUST

    Graduated top of the programme with the Gold Medal, CGPA 4.0 / 4.0. Before that, BS Computer Science at the University of Gujrat with the Top Project award.

Career highlights

  1. 21.8×

    Inference optimization

    Cut p95 latency in production from 27.2 seconds to 1.25.

  2. 8 Million

    Predictions a day

    20 vision models on shared GPUs, serving 500+ devices in the field.

  3. 88%

    Less GPU memory

    Cut peak VRAM on the same workload and ended a recurring crash.

  4. $50K

    Saved every year

    One shared serving layer for over 3 million requests a month, on far less cloud compute.

  5. 450 Million

    Pixels per image

    AI detection on satellite images, each the size of 217 Full HD screens.