Experience & companies
Professional experience.
My roles, responsibilities, and projects in one place. Choose a company to explore the work in detail.
Enhanced Robotics (Tenniix Official)
Algorithm Engineer
Dec 2025 – Present Shenzhen, China
Own the visual-perception lifecycle for real-time tennis AI systems.
Responsibilities & contributions
- Built temporal ball and person-attribute models, court keypoint detection, and dual-camera geometric fusion.
- Developed auto-labeling and model export workflows for edge deployment and continued data iteration.
Projects at Enhanced Robotics
15 projectsMultilingual TC-ResNet Keyword Spotting on ESP32-S3
Trained and deployed seven-language TC-ResNet14 command models with 2–3-second audio windows and a shared TC-ResNet8 wake-word model with a 1-second window on ESP32-S3, using an INMP441 microphone.
View project detailsMultilingual KWS Dataset Generation with Voice Design and Cloning
A completed synthetic speech data pipeline for keyword spotting across seven languages, combining voice design, identity-preserving cloning, ASR verification, and recoverable dataset assembly.
View project detailsPerson Pose Estimation with Distance and Racket-Hand Recognition
A single-image YOLOv8 pose model combining person boxes, 17 body keypoints, camera-depth regression, and racket-hand recognition, with pose-derived free-hand gestures and RDK X5 export and inference.
View project detailsDual-Camera 3D Tennis Scene Understanding
A calibrated HD/4K tennis-perception pipeline that combines court pose, multi-branch temporal ball detection, size-derived ball depth, and person-depth predictions in shared 3D court coordinates.
View project detailsGrid-Guided Tennis Court Keypoint Detection
A six-channel court detector that pairs RGB with a separate synthetic grid, predicts line–grid intersection landmarks, and reconstructs court lines for camera-pose estimation.
View project detailsPerson Detection with Distance, Gesture, and Player Status Prediction
A single-image YOLO model with person boxes, a camera-depth regression head, and independent hand-raise and player-role classifiers, supported by reviewed teacher labels and edge export.
View project detailsSAM-3D-Body Auto-Labeling for Person Detection
A SAM-3D-Body labeling workflow that produces person boxes, camera-depth estimates and hand-raise labels, followed by ReID-assisted role review, outlier handling and YOLO export.
View project detailsTemporal Tennis Ball Detection with Ball Diameter Prediction
A three-frame ball detector with explicit diameter supervision, size-aware postprocessing, reviewed temporal datasets, and ONNX/TorchScript deployment paths.
View project detailsUnified Temporal Ball Detection & Ball-Size Regression
An extension of the three-frame ball detector with a dedicated distribution-regression branch that learns ball size alongside localization and frame identity.
View project detailsTennis Court Keypoint Localization System
A two-stage court-localization pipeline that detects seven court and net-post landmarks, then refines line and post cues in local crops for downstream camera geometry.
View project detailsTwo-Frame Temporal Player Detection System
A six-channel person detector that combines a current frame with a configurable earlier background frame, using matched crops and optional camera-motion compensation.
View project detailsVision-Based Player Gesture Safety Control System
A raised-hand gesture component for tennis-machine stop control, first implemented separately and later consolidated into the shared person-attribute detector.
View project detailsCropped-Ball Radius Estimation for Tennis Depth Inference
A dedicated CNN that estimates ball radius in crop-space pixels using distribution regression, with single-frame and paired-frame inputs for downstream camera-geometry calculations.
View project detailsInteractive SAM3 Auto-Labeling for Tennis Ball Detection
A PyQt/SAM3 annotation workflow with multi-frame prompts, mask propagation, review and correction loops, and ISAT export in the original image coordinates.
View project detailsTemporal 3-Frame Tennis Ball Detection System
A YOLO-based temporal detector that stacks three RGB frames into nine channels and returns frame-indexed ball detections, with full-frame and center-crop inference branches.
View project details
Benign Innovations (Co-founding Startup)
Algorithm Engineer
Oct 2024 – Dec 2025 Shenzhen, China
Led applied perception R&D for tennis analytics and autonomous lawn-care robotics.
Responsibilities & contributions
- Integrated court geometry, ball and player tracking, pose, and ReID into an on-device tennis analytics stack.
- Built auto-labeling, segmentation, and deployment pipelines for lawn-care robot perception.
Projects at Benign Innovations
9 projectsReal-Time Tennis Analytics with Multi-Modal Vision on Edge Devices
A real-time, edge-powered tennis analytics stack combining player tracking, pose, ball trajectory, and court mapping.
View project detailsGround Segmentation for Obstacle Avoidance
A point-cloud processing pipeline for ground segmentation and obstacle extraction in outdoor robot navigation.
View project detailsTennis-Court Simulation in Webots and Isaac Sim
Synthetic tennis-court simulation work in Isaac Sim and Webots for generating ball-tracking datasets and testing camera layouts.
View project detailsAuto-Labeled Weed Detection and Species Classification
A lawn-care perception component that distills foundation-model masks and classifications into a compact weed detector.
View project detailsSprayer-Robot Detection from Auto-Labeling to Deployment
An object detection pipeline for a sprayer robot, from foundation-model auto-labeling to YOLOv11n edge deployment.
View project detailsGrass Quality Segmentation
A grass quality segmentation system that converts weak labels into patch-level supervision and produces quality heatmaps.
View project detailsCharging-Station Keypoints and 3D Localization
A lightweight charging-station keypoint detector for 3D localization and auto-parking.
View project detailsStereo Matching from Auto-Labeling to Edge Deployment
A stereo matching pipeline that uses a large teacher model for auto-labeling and a smaller student model for edge deployment.
View project detailsInstance Segmentation from Auto-Labeling to Edge Deployment
An end-to-end instance-segmentation workflow spanning foundation-model auto-labeling, training, evaluation, and edge deployment.
View project details
LinxAI Intelligent Technology Co., Ltd.
Algorithm Engineer
Dec 2023 – Oct 2024 Shenzhen, China
Developed deep-reinforcement-learning locomotion policies and simulation-to-robot transfer workflows.
Responsibilities & contributions
- Trained proprioception-only locomotion policies in Isaac Gym and transferred them to a custom quadruped.
- Developed flat- and rough-terrain policies; published a physical-robot demonstration including 15 cm stairs.
Projects at LinxAI
3 projectsHybrid Deep RL and MPC for Quadruped Locomotion
A hierarchical quadruped locomotion framework that combines Deep Reinforcement Learning with Model Predictive Control.
View project detailsDeep Reinforcement Learning for Custom Quadruped Locomotion
Proprioception-only locomotion policies transferred from simulation to a custom quadruped across flat and rough terrain.
View project detailsDeep RL Locomotion on the Unitree A1
Deep reinforcement learning experiments for quadruped locomotion on a Unitree A1 robot.
View project details
XPENG Robotics
Deep Learning Engineer
May 2021 – Oct 2023 Shenzhen, China
Built and optimized perception workflows across indoor 3D vision, robot interaction, and autonomous systems.
Responsibilities & contributions
- Developed depth, segmentation, human perception, and robot-interaction models.
- Built a camera–LiDAR fusion pipeline that increased depth-label density from 6% to 58% using 60 scans.
Projects at XPENG Robotics
15 projectsDoor and Window Keypoint Detection
A keypoint detection system for indoor doors and windows, designed to improve structural perception in indoor robotics environments.
View project detailsRobot-Arm Segmentation for Background Filtering
A binary semantic segmentation pipeline for separating a robotic arm from the background.
View project detailsAuto-Labeling for 6-DoF Grasp-Pose Detection
An RGB-D auto-labeling pipeline that generates 6-DoF grasp-pose annotations for robotic pick-and-place in unseen domestic environments.
View project detailsFeature Matching for Visual Odometry and SLAM
A survey and comparison of feature matching methods for visual odometry, visual localization, and SLAM.
View project detailsHuman–Object Interaction on Custom Data
A custom human-object interaction dataset and model training workflow.
View project details3D Multi-Person Pose Estimation
A survey of 3D multi-person pose estimation methods for recovering human pose in both relative and absolute coordinate settings.
View project detailsFacial Expression Recognition (FER)
A survey of recent facial expression recognition algorithms and emotion categories.
View project detailsHuman Mesh Recovery with Parametric Body Models
A survey of human mesh reconstruction methods using parametric body models such as SMPL.
View project detailsSpatio-Temporal Human Action Recognition and Localization
A spatio-temporal action recognition project using contextual relationships between actions, people, and surrounding objects.
View project detailsDense Depth Labels from LiDAR–Camera Fusion
A synchronized LiDAR and multi-view pipeline that increased depth-label density from 6% to 58% through calibrated frame fusion.
View project detailsSelf-Supervised Multi-View Stereo Reconstruction
A self-supervised multi-view stereo reconstruction workflow for generating stronger depth supervision on custom data.
View project detailsUncertainty Modeling for Stereo Depth Estimation
A stereo depth estimation project focused on modeling both aleatoric and epistemic uncertainty for safer real-time depth prediction.
View project detailsReal-time Aleatoric Uncertainty Estimation in Semantic Segmentation
Added real-time aleatoric uncertainty estimation to a semantic segmentation pipeline so uncertain regions can be filtered before downstream use.
View project detailsIndoor Semantic Segmentation
An indoor semantic segmentation project comparing transformer-based and CNN-based segmentation models on custom robotics data.
View project details3D Object Detection on Indoor Environments
An indoor 3D perception pipeline combining stereo depth, 3D object detection, and 6D pose estimation.
View project details
Southern University of Science and Technology
Research Assistant
Nov 2020 – Apr 2021 Shenzhen, China
Researched 3D scene parsing across semantic mesh segmentation and point-cloud object detection.
Responsibilities & contributions
- Worked on semantic mesh segmentation, mesh preparation, and point-cloud object detection.
ROPEOK Technology Group
Algorithm Engineer
Oct 2019 – Sep 2020 Xiamen, China
Developed real-time person re-identification and attribute-recognition systems for multi-camera environments.
Responsibilities & contributions
- Developed person re-identification across six cameras and human-attribute recognition.
Projects at ROPEOK
2 projectsHuman Attributes Recognition System
A human attributes recognition system focused on safety-related visual attributes and edge deployment.
View project detailsReal-Time Person Re-Identification Across Six Cameras
A real-time person re-identification system for intelligent video surveillance across six adjacent-street cameras.
View project details
Pakistan Aeronautical Complex
Design Engineer
Jan 2017 – Aug 2017 Kamra, Attock, Pakistan
Contributed to smart-display graphics and EEG feature-extraction initiatives.
Responsibilities & contributions
- Worked on OpenGL display software and EEG signal feature extraction.