Hanlin Liu Email and Phone Number

● Utilize Graph Mining techniques in ML pipelines to compute feature similarity scores for all Ads publisher accounts. Detect duplicate accounts and shorten the pipeline running time by 85% by adopting the online Pairwise approach.● Build and optimize Graph Neural Network model that fights Ad spam, catching $XXM per year with 95%+ precision.● Design and implement ML infrastructure and launch continuous re-computation pipelines in an optimal patten to detect identity-changing spammers.● Improve the ML inference framework for both approval system and throttler system to reduce model inference latency and increase productivity.

● Develop C++ SDK (dlls) to perform conversions from TensorFlow saved model to ONNX model and to TensorRT model on Windows edge devices with Nvidia GPU and run inferences with the implemented APIs.● Perform graph surgeon on ONNX models to remove the top Filter Detection layer (nms layer) for the object-detection models and add it back in the post-processing after doing inference using TensorRT engine.● Develop C++ applications to run image preprocessing and perform inferences by implementing TensorFlow C APIs to meet customers’ latency requirements. The latency using such C++ applications is reduced by over 5 times in comparison to the Python counterparts.● Design and develop software stack on our LandingLens Software Platform for TensorRT support using Docker.

● Lead a software team to work on the AGV Assignment Service development. Be in charge of designing the software architecture, developing APIs/submodules, implementing and optimizing algorithms and logics to increase AGV utilization and the robotic system efficiency.● Collect functional requirements from testing and operation teams, transform the needs to service or logic development items, review task distribution and level of effort (LOE) assessment with the team, start development with the team, discuss the data structures and algorithms for optimization, perform code reviews and test all the use cases and deploy to production.● Design, develop and optimize the logics and algorithms of robot dispatch and traffic management system including decision making of section selection, AGV trajectory planning and rerouting logic using A* algorithm, intelligent path reserving logic, AGV bypassing logic, AGV charging logic, etc. using C++11 and deploy to Linux environments.● Develop an end-to-end AGV Travel Time Estimator (TTE) using Machine Learning with Python 3 TensorFlow, perform data exploration/preprocessing, feature engineering, design and develop and train a Machine Learning model by implementing Deep Neural Networks to predict the AGV travel time. Perform hyperparameter tuning to optimize the model, increase accuracy and reduce RMSE. Deploy to production using Python Flask.● Implement Redis cache system in the AGV Assignment Service to update AGV position in real-time to meet the read-heavy requirement and prevent stale vehicle position data.

● Design and implement robust and scalable software solutions to support production operations, improve robotic system throughput and meet business requirements.● Extensively apply various C++ data structures in software developments including set, map, stack, queue, heap, BST and implement algorithms using BFS, DFS, recursion, backtracking, divide and conquer, dynamic programming etc.● Lead the robotics project in Staples Charlotte Fulfillment Center, including robotics layout design, software development and deployment, system performance analysis and feature enhancements to support highly efficient operation and meet throughput requirements.● Design and develop AGV dispatch and traffic management software system, deploy and run the software package in Linux environments. Debug and analyze heap and stack information using Linux GDB when core dump generates.● Design database schema in PostgreSQL and troubleshoot current bugs in database functions, increase the productivity and improve the robustness and efficiency of our robotic system.● Quickly diagnose and troubleshoot errors that occur in both live production environments and testing environments.● Work closely with multiple cross functional engineers to bring an idea from concept design to development, improvement and final realization.

● Implement closed-loop control for the autonomous, free-swimming underwater robot.● Establish software identification to reconstruct 3D kinematics and set up large data storage system.● Achieve integration and synchronization of multiple high-speed cameras with volumetric particle image velocimetry (V3V PIV) system.● Establish software identification to reconstruct 3D kinematics and set up large data storage system.● Instruct a graduate-level course: Engineering Data Analysis.

● Research on fin ray stiffness and fin morphology on ribbon-fin-based propulsion by developing and testing a robotic ribbon fin model.● Designed and developed software system for a self-contained, free-swimming robot propelled by an undulatory fin. The robotic system features a compact slender 3D-printed hull with 16 Maxon DC motors, 2 Li-Ion batteries and 5 custom designed PCBs encapsulated inside. Designed the robotic vessel using Solidworks and applied a 3D printer to manufacture the hull sections. Designed and developed software system in C for the underwater vehicle to achieve various motion patterns including forward swimming, rapid brake, backward swimming, vertical swimming and station keeping in the water. Deployed the software to 9 Microchip dsPIC33EP512GM304 (16-bit) microcontrollers, implemented communication layers to establish a Comms Bus (through UART module) and a Sync Bus between 1 master MCU and 8 slave MCUs. Implemented PID algorithm to control 16 motors (through PWM peripheral) and utilized current sensors as well as voltage sensors to measure power consumption of motors. Applied a Zigbee RF module to communicate with the host computer and wirelessly control the robot via radio signals. Incorporated various sensors (IMU, pressure sensor, temperature sensor, leakage sensor) in the electronic system for orientation measurement, pressure/temperature monitoring and leakage detection (through I2C, ADC).● Investigation on the maneuvering characteristics and propulsive performance of the free-swimming robotic platform propelled by an undulatory fin.● Research on the flow patterns and vortex structures of free-swimming undulatory-fin-based robot using volumetric PIV system.

● Design, develop and implement a software system for a real-time dynamic positioning simulator using C.● Investigation on dynamic positioning system of multi-vessels arranged side-by-side● Study on the effect of heel and trim angles on a semi-submersible drilling platform with dynamic positioning system -Performed a numerical study to simulate a semi-submersible platform with dynamic positioning system (DPS) under various heel and trim angles by considering different thruster-thruster interaction patterns. -Achieved the time evaluation results of positioning accuracy and power consumption. Established a dynamic positioning capability diagram to analyze its positioning performance.● Research on the effect of thruster failure on the optimal operational direction of DP-2 semi-submersible platform -Led our team to conduct model tests of the DPS on a semi-submersible platform for evaluating its positioning capability with different thruster failure patterns in the presence of winds, waves and ocean currents. -Conducted numerical simulations of the corresponding experimental cases. Analyzed the numerical results in comparison with the experimental results at various heading angles to obtain an optimal operation direction.● Conducted numerical simulation and model experiments to investigate the effect of thruster failure patterns on the optimal operation direction of a DP-2 semi-submersible platform.