Jiarui Chen

I am a Master of Science student in Computer Science at the University of California San Diego. I received my Bachelor of Computing in Computer Science from the National University of Singapore.

I have experience in machine learning systems, backend infrastructure, and applied research. My work includes time-series forecasting for observability platforms, retrieval-augmented GPT tools, Spark-based ML pipelines, and VR/robotics simulation projects.

 /  GitHub  /  LinkedIn  /  Project Demos

Experience

Research Assistant

• Developed a VR-based surgical training simulator for breast cryoablation using Unreal Engine, providing a repeatable interactive alternative to traditional lecture-based or physical phantom training.
• Estimated a $300+ per-trainee training cost reduction by replacing single-use physical phantom models with repeatable VR simulation. Collaborated with the National University Hospital to evaluate the simulator with 21 medical student participants.
• Implemented an in-game tablet for surgery planning, 2D ultrasound cross-sectional imaging, cryoablation simulation, and a procedural point system that tracks surgical accuracy.

Machine Learning Engineer Intern

• Developed an end-to-end time-series data forecasting & anomaly detection pipeline with deep learning models and algorithms, later deployed on JNJ’s Kubernetes GPU cluster after containerized as Docker image.
• Exposed the service as an internal Flask API integrated into the J&J Grafana workflows and used by ~500 engineers for real-time monitoring. Enabled access to expected patterns, capacity trends, faster incident triage, and proactive scaling.
• Built a reusable data preparation workflow that queried observability metrics from VictoriaMetrics and converted them into model-ready time-series datasets for downstream forecasting and anomaly detection services.
• Implemented forecasting in PyTorch, progressing from an autoregressive LSTM baseline to Temporal Fusion Transformer (TFT) with seasonal/weekly temporal feature engineering, achieving average estimation error within 3.3% for metrics on a 100% scale.
• Led a team of 5 interns to develop an internal GPT-based advisory chatbot for sales reps. Built the retrieval core by embedding user queries and matching them against domain-specific knowledge with cosine similarity search.
• Integrated retrieved internal content into GPT-based response generation to deliver more grounded, context-aware product guidance, improving sales reps’ ability to prepare for client visits.

Back-end Web Developer

• Deployed Master-Slave control clusters for Kylinsec's cloud operating system. Containerized environments using Docker within Jenkins-based CI/CD pipeline.
• Fixed a CPU core allocation bug in the Zombie Cloud module, restoring up to 2% cluster capacity through optimizing scheduler and query logic.

Publications

• J. Chen, et al. Enhancing Breast Cancer Cryoablation Training via VR Simulation, BMC Medical Education, under review, 2026.
• J. Chen, et al. Private Chat in a Public Space of Metaverse Systems, arXiv:2511.07993, 2025.
• J. Chen. Memory Assisted LLM for Personalized Recommendation System, arXiv:2505.03824, 2025.

Projects

Autonomous Mobile Robot

Completed a five-part robotics project series on the RB5/mBot rover platform, covering the full autonomy stack. Built an open-loop waypoint controller with a kinematic bicycle model. Closed the loop with AprilTag-based visual servoing and camera calibration. Implemented an EKF-SLAM system to simultaneously localize the robot and map unknown landmarks in a 10×10 ft arena. Designed and compared two path planners—one optimizing for safety margin and another for minimum distance—with obstacle avoidance. Integrated all subsystems into a coverage-based “Roomba” navigator using a subsumption architecture.

Memory Assisted LLM for Personalized Recommendation System

Developed an LLM-based recommendation system with ‘memory’ to improve recommendation precision using personalized history. Conducted extensive experiments on movie rating and cross-domain prediction with MovieLens 100K and Amazon Review datasets. Improved memory efficiency by integrating external knowledge sources.

Private Chat Channels in Casual Social VR Systems

Implemented a private communication channel packet with Unity under supervision by Prof. Anand Bhojan. Analyzed APIs of major VR platforms and built a world instance enabling private chat in VRChat. Formulated and tested four research questions to enhance user experience with 21 participants.

PPP Loan Risk Prediction with Spark

Built a Spark-based ML pipeline on Databricks to process large-scale PPP loan data, including class rebalancing, missing-value imputation, feature selection, and train/test split preparation. Trained and evaluated a Random Forest classifier using PySpark ML for binary prediction of full repayment versus charge-off on real-world loan records.

The Beyonders – Turn-Based Strategy RPG

Co-developed a turn-based, grid-based tactical RPG in Unreal Engine 5.3 inspired by the Chinese web novel Lord of the Mysteries. Designed strategic combat with six playable Beyonder characters, each with unique abilities tied to the “Sequences” power system. Implemented party management, a multi-level campaign (tutorial, Church, and Courtyard stages), and a complete game guide. Contributed to gameplay programming, visual effects, and level design across a five-person team.

Photo De-hazing via Density and Depth Decomposition

Studied and reproduced the CVPR 2022 paper “Self-augmented Unpaired Image Dehazing via Density and Depth Decomposition” (D4) by Yang et al. under the mentorship and supervision by Prof. Huailiang Tan. Gained hands-on experience with self-augmented unpaired image dehazing, GPU-accelerated training via CUDA, and deep learning-based image restoration techniques. Credit to the original D4 repository by YaN9-Y.

TeamBuilder Desktop Application

Led a 4-person team to build a JavaFX contact & team-formation manager for students. Designed features for searching, tagging, and skill-based grouping using Object-Oriented Design. Implemented core MVC architecture improvements and persistent JSON data storage, enabling users to form balanced teams efficiently.

Space-Lin Task Tracker

A 1.6k LoC greenfield individual project for a software development module at NUS. Developed a desktop task-tracking application using JavaFX, incorporating I/O redirection techniques for text UI testing and Gradle build integration with JUnit tests.

Skills

Languages: Python, Java, C++, C, C#, SQL, JavaScript

ML / Data: PyTorch, TensorFlow, Keras, Scikit-learn, CUDA, Spark, Hadoop, Databricks, PyTorch Forecasting

Infrastructure: Docker, Kubernetes, Jenkins, Grafana, Prometheus, VictoriaMetrics, PostgreSQL, Flask

Graphics / XR: Unity, Unreal Engine, OpenGL, ROS2

Tools: Git, Vim, JUnit5, JavaFX, Grails

Education

University of California San Diego

National University of Singapore

Awards & Certificates

• 2025 — Certificate of Distinction, Computer Graphics and Games Focus Area, NUS
• 2025 — Certificate of Merit, Database Systems Focus Area, NUS
• 2025 — 26th STePS NUS School of Computing Term Project Showcase, Third Prize
• 2024 — 25th STePS NUS School of Computing Term Project Showcase, Best Project Award – Platinum
• 2024 — National University of Singapore, Dean's List

Design and source code from Jon Barron's website. Template by Leonid Keselman.