Professional Work

Where I've worked, what I shipped, and what I learned doing it.

  1. Vizgard Ltd

    Jul 2021 – Aug 2024

    Founding Software Engineer · London, United Kingdom

    Joined as one of the earliest engineers and helped take the company from research prototype to a venture-funded computer vision platform. Owned the real-time CV pipeline end-to-end — from camera ingest through model inference to WebRTC delivery.

    • →Architected an event-driven, lock-free MPMC double-buffer pipeline that hit 6 simultaneous 1080p streams on a single NVIDIA Jetson and 40 streams across 4× Ada RTX 6000.
    • →Cut critical-path latency by ~3× (measured in Nsight Systems / Remotery) by moving CPU stages into custom CUDA kernels with no loss of model accuracy.
    • →Trained and deployed object detection (DETR/YOLO), tracking (SiameseRPN++, DeepSORT), pose estimation (AlphaPose), and face redaction — PyTorch/TensorFlow on GCP, TensorRT in production.
    • →Built the low-latency WebRTC streaming server (GStreamer + Node.js) that delivers HD video to browsers and RTSP endpoints.
    • →Contributed to the company's growth from early prototype to securing over $2.5M in venture funding.
    • C++
    • CUDA
    • TensorRT
    • PyTorch
    • GStreamer
    • WebRTC
    • Node.js
    • Jetson
    • GCP

  2. VinAI Research

    Jan 2020 – Jul 2021

    Software Research Engineer

    Research engineer on the face analytics team at VinAI (later acquired by Qualcomm Research). Focused on the production gap between research models and on-device inference.

    • →Designed a two-stage infrared anti-spoofing system, achieving < 3% false acceptance rate (FAR).
    • →Optimized the face recognition model via knowledge distillation and network pruning — 4× faster at equivalent accuracy.
    • Python
    • PyTorch
    • TensorFlow
    • Knowledge Distillation
    • ONNX