Education
MSc in Electrical Engineering, University of Southern California, California, US, Aug 2023 - Dec 2024 | GPA: 3.94/4.0
BSc in Internet and Multimedia Technologies, The Hong Kong Polytechnic University, Hong Kong, Aug 2019 - May 2023 | First Class Honors | GPA: 3.61/4.0
Technical Skills
Languages
Python, Node.js, C++, Java, C#, JavaScript, HTML, PHP
ML & CV
PyTorch, TensorFlow, Scikit-learn, OpenCV, computer vision, spatio-temporal modeling, graph neural networks, deep learning, machine learning
LLM & AI Agent
RAG, prompt engineering, context structuring, retrieval refinement, LLM fine-tuning
Backend & Deployment
AWS, Docker, Flask/FastAPI, REST APIs, Unity, ARFoundation, SQLite/MySQL, NoSQL
Work Experience
Project Associate | The Hong Kong Polytechnic University
Jan 2025 – Present- Owned end-to-end delivery of a patented AI-assisted AR hand-hygiene application, spanning data pipelines, model integration, backend services, and production deployment across clinical training environments.
- Independently shipped core functionality in one month, replacing an estimated ~6 months of outsourced engineering work and accelerating stakeholder validation for product rollout.
- Built a multi-stage ML system for data collection, preprocessing, training, optimization, and cross-platform inference, supporting mobile, desktop, edge devices, and AR glasses in both online and offline modes.
- Scaled the solution to 20+ hospitals in Hong Kong; supported institutional adoption, procurement discussions, and external presentations at ICPIC, IICC, and Hospital Authority Convention.
- Stack: Python, PyTorch, TensorFlow, OpenCV, AWS, Unity, C#, ARFoundation, Docker, REST APIs, SQLite/MySQL
Research Assistant (Part-time) | The Hong Kong Polytechnic University
Jan 2022 – Oct 2024- Designed and deployed an AI platform for high-resolution indoor contact mapping, covering data engineering, detection/tracking pipelines, cloud services, and analytics workflows for community-scale operations.
- Built production-oriented CV and spatio-temporal modeling pipelines that generated contact graphs and risk signals for public-health decision support in Hong Kong deployment settings.
- Modeled relative transmission risks across indirect person-to-site and direct person-to-person contact pathways, translating research outputs into interpretable operational insights.
- Methods/Tools: spatio-temporal modeling, multi-object tracking, risk scoring, privacy-aware data processing. Python, PyTorch, Flask/FastAPI, AWS
Project Experience
HealthAgent and Structured EHR Graph System
Self-Directed LLM Systems / AI Agent Engineering · 2025 – Present- Built a local health-consultation AI agent with FastAPI, React, and SQLite/PostgreSQL support, including multi-user profiles, persistent sessions, doctor-style multi-turn intake, exportable reports, and graph-based association analysis.
- Designed a structured EHR backend centered on relational clinical entities and a unified graph-edge model, enabling LLM tool calling over patients, encounters, conditions, observations, medications, reports, and derived associations.
- Implemented graph-query and record-resolution workflows for summary, timeline, active problems, lab trends, neighbor lookup, path search, and rare-association retrieval, fetching only task-relevant context instead of full patient-history prompts.
- Organized prompt packs and provider routing for local agent execution, supporting Codex CLI by default with HTTP API fallback, while keeping prompts modular and user context intentionally minimal for safer and more token-efficient interactions.
Impact of Optimization Algorithm on Neural Network Generalization
USC Viterbi School of Engineering · Sep 2024 - Dec 2024- Conducted a controlled study on how optimization algorithms (SGD, Adam, L-BFGS) influence generalization on CIFAR-10, standardizing model architecture, data pipeline, and training budget to enable fair comparisons.
- Built a fully reproducible experimental workflow (data loading, training, evaluation, logging, and visualization), and reported results across test accuracy, convergence speed, stability, and compute cost.
- Performed robustness-oriented analyses by probing sensitivity to hyperparameters and training noise, highlighting failure modes and practical tuning considerations.
- Quantified trade-offs between first-order and quasi-second-order methods, showing first-order optimizers generally provide a stronger balance of training efficiency and noise robustness under comparable compute constraints.
TrojanMap Project
USC Viterbi School of Engineering · Jan 2024 - May 2024- Developed a geographic information system in C++ to model and query a map of USC and its surroundings.
- Implemented core data structures (Node and TrojanMap classes) to represent locations and paths; added auto-completion, category-based search, and location queries.
- Implemented shortest-path solvers (Dijkstra, Bellman-Ford) for efficient real-time navigation queries.
- Solved the Traveling Salesman Problem (TSP) via brute force, backtracking, and 2-opt optimization, reducing path computation complexity.
Engine Sound Detection Based on Deep Learning
The Hong Kong Polytechnic University · Aug 2022 - May 2023- Developed deep-learning models for engine sound event detection and correlation; achieved 90% accuracy identifying specific engine issues and reduced maintenance response time by 25%.
- Used MFCC feature extraction to improve pattern discrimination; increased detection precision by 15%.
- Trained and compared L3-net variants with ConvNet, ResNet, and DenseNet baselines; found ResNet improved accuracy by 5% on complex sound recognition.
- Applied transfer learning with pre-trained models, reducing training/testing time by 50% and accelerating iteration.
Dungeon Mission Game (Unity, C#)
The Hong Kong Polytechnic University · Jan 2022 - May 2022- Co-developed a browser-based roguelike game in Unity, combining procedural generation and AI mechanics for a dynamic player experience.
- Implemented multiplayer using Photon PUN2, enabling up to 50 concurrent players with real-time interaction.
- Engineered enemy AI behaviors including two unique bosses; contributed to a 40% increase in engagement and 25% improvement in positive reviews.
- Published at simmer.io/@MENG/team-04-dungeon-mission (100+ players).
Publications & Patent
Publication: Jiahui Meng, Justina Liu, Lin Yang*, et al. An AI-empowered Indoor Digital Contact Tracing System for COVID-19 Outbreaks in Residential Care Homes. Infectious Disease Modelling, 2024. doi:10.1016/j.idm.2024.02.002.
- Patent: China Short-term patent No. 202510682954.3 (Jun 2025)
- Finalist (Prix Hubert Tour - Innovation Academy), Oral Presentation, 8th International Conference on Prevention & Infection Control (ICPIC), Geneva, 2025
- Oral Presentation, 9th International Infection Control Conference (IICC), Hong Kong, 2025
- Booth Presentation, Hospital Authority Convention, Hong Kong, 2025
Honors & Awards
- PolyU Junior Researcher Mentoring Programme (2025)
- Dean's Honor List of Outstanding Students, The Hong Kong Polytechnic University (2020 - 2021)
Leadership
The Hong Kong Polytechnic University Golden Z Club Hong Kong - Promotion Secretary (2020 - 2021)
- Organized a sign language workshop (50+ participants) and an interview workshop (60+ attendees); improved engagement by 40%.
- Initiated mentoring programs for 50+ South Asian children, improving academic performance by 20%; organized volunteer education activities impacting 100+ children.
- Coordinated social activities and non-profit events with team members.