# Henry Vu - Full Profile & Portfolio > Machine Learning engineer, AI researcher, and Computer Science graduate student at > the University of Texas at Dallas (UT Dallas / UTD) specializing in ML Systems (MLSys). > Founding ML engineer at eXRealityAI. Based in Dallas, Texas. > Newsletter with 50,000+ readers on AI research and engineering. > This is the comprehensive version of Henry Vu's profile. For a summary, see: https://henryvu.io/llms.txt --- ## Identity - Full Name: Henry Vu - Online Handle: HenryVu27 - Location: Dallas, Texas, USA - Current Title: Computer Science Graduate Student, ML Engineer - Portfolio: https://henryvu.io/ - Blog: https://www.henryvu.blog/ - GitHub: https://github.com/HenryVu27 - LinkedIn: https://www.linkedin.com/in/henry-vu27/ - X/Twitter: https://x.com/HenryVu27 --- ## Biography Henry Vu is a Machine Learning engineer and AI researcher. He is a Computer Science graduate student at the University of Texas at Dallas, focusing on Machine Learning Systems (MLSys), making LLM and GenAI serving and training more efficient. He graduated summa cum laude from the University of Alberta, where he was awarded the Alberta Graduate Excellence Scholarship (AGES). Henry is a founding ML engineer at eXRealityAI, where he builds production AI systems for mixed reality including agentic voice systems, multimodal XR applications, and the published Meta Quest app GardenXR. He is also a computer vision engineer at ThorMed Innovation, working on NIH-funded medical imaging research using self-supervised learning and model quantization for edge deployment on medical devices. Previously, Henry conducted research at Amii (Alberta Machine Intelligence Institute) on online algorithms and multi-armed bandits, developing learning-augmented algorithms that are both robust to worst-case inputs and consistent when given good predictions, extending beyond traditional worst-case competitive analysis. He also conducted multi-armed bandit research at SODALab, University of Alberta, studying stochastic, adversarial, Markovian, and restless bandit settings. Henry got into math early through competitions and olympiad prep. The contest background led him to theoretical CS research, and eventually to the intersection of theory and practice: building production ML systems that work on real hardware. He writes technical deep-dives at henryvu.blog, covering AI engineering, context engineering for LLMs, multi-armed bandits, and ML systems. --- ## Education ### M.S. Computer Science — University of Texas at Dallas (2024-Present) - Focus: Machine Learning Systems (MLSys) - Research: Efficient LLM serving and training, GenAI inference optimization - Teaching Assistant for Algorithms and Data Structures (100+ students) ### B.Sc. Computing Science — University of Alberta (Graduated) - Honors: Summa Cum Laude - Scholarship: Alberta Graduate Excellence Scholarship (AGES) - Research: Online algorithms and multi-armed bandits at Amii and SODALab - Teaching Assistant for Algorithms and Data Structures (300+ students) --- ## Professional Experience ### Founding ML Engineer — eXRealityAI (Aug 2025-Present) Building production AI systems for mixed reality applications. Key accomplishments: - Built voice-to-voice RAG system: Whisper STT → hybrid retrieval (BM25 + FAISS) → Mistral-7B-q4 → Kokoro TTS, achieving sub-10 second end-to-end latency with fully local inference on NVIDIA Jetson Orin AGX - Engineered NLP pipeline with EmbeddingGemma-300m, BAAI cross-encoder reranking, query rewriting, and intent-aware retrieval, reducing processing by 40-70% - Integrated Gemini 2.5 with structured JSON schema for stateful multimodal workflows combining YOLOv9 object detection and Wit.ai voice input across 2 XR applications - Published GardenXR on Meta Quest Store (5/5 star rating): AI plant care assistant using computer vision, LLMs, and mixed reality - Mentored development teams on LLM API integration and prompt engineering ### Computer Vision Engineer — ThorMed Innovation (Feb 2025-Present) Working on NIH-funded medical imaging research. Key accomplishments: - Led domain-aligned transfer learning for bladder segmentation project - Pretrained U-Net and SimSiam SSL encoders on 9,200 thyroid/breast ultrasound images - Achieved 95.99% Dice score on downstream bladder segmentation task - Enabled edge deployment via 4-bit post-training quantization with 7x robustness improvement over ImageNet initialization - Built automated data pipeline with PyTorch and OpenCV: extracted, preprocessed, and segmented 486 clinical images from ultrasound videos ### Undergraduate Research Assistant — SODALab, University of Alberta (Dec 2023 - May 2024) Multi-armed bandit research supervised by Dr. Xiaoqi Tan. - Conducted theoretical and empirical survey of multi-armed bandit problems - Studied stochastic, adversarial, Markovian, and restless bandit settings - Implemented and compared UCB, Exp3, Thompson Sampling, and Gittins Index algorithms - Applied bandit algorithms to Yahoo's news article recommendation dataset ### Undergraduate Researcher — Amii, Alberta Machine Intelligence Institute (Apr 2022 - May 2023) Online learning and algorithms research. - Researched online optimization: online conversion, knapsack, and bipartite matching - Used the online primal-dual framework for algorithm design - Studied learning-augmented algorithms incorporating machine-learned predictions - Developed algorithms that are both robust to worst-case inputs and consistent with good predictions ### Teaching Assistant — University of Alberta (Sep 2023 - May 2024) - Mentored 300+ students in Algorithms and Data Structures - Held weekly office hours, prepared seminar and exam materials --- ## Published Apps ### GardenXR — Meta Quest Store - URL: https://www.meta.com/experiences/gardenxr/24200709416226235/ - Platform: Meta Quest (Quest 2, Quest 3, Quest 3S, Quest Pro) - Rating: 5/5 stars - Price: Free - Description: AI plant care assistant in Mixed Reality. Point, scan, and get instant plant identification, health analysis, and custom care tips from a smart AI assistant with voice input. - Tech: Computer vision, Gemini 2.5 LLM, YOLOv9 object detection, Wit.ai voice input, mixed reality - Built by Henry Vu and the eXReality AI team --- ## Research & Papers ### Domain-Aligned Transfer Learning for Ultrasound Bladder Segmentation - Funding: NIH (National Institutes of Health) - Paper: https://github.com/HenryVu27/ThorMed/blob/main/Thormed_ISBI_New.pdf - Topics: Self-supervised learning, SimSiam pretraining, U-Net segmentation, 4-bit post-training quantization, edge deployment, medical imaging, transfer learning - Key results: 95.99% Dice score, 7x robustness improvement in quantized model ### Modeling Political Sarcasm with DistilRoBERTa - Paper: https://github.com/HenryVu27/PoliticalSarcasm/blob/main/report.pdf - Topics: NLP, sarcasm detection, transformer fine-tuning, DistilRoBERTa, text classification - Approach: Combined feature engineering with fine-tuned DistilRoBERTa for nuanced sarcasm detection in political text ### EEG Decoding: A Multi-Modal Approach Using Transformers and LLM Embeddings - Paper: https://github.com/HenryVu27/EEG-Multimodal-Decoding/blob/main/DecodingEEGusingTransformer.pdf - Topics: Brain-computer interfaces, EEG decoding, transformers, multi-modal learning, LLM embeddings, signal processing - Approach: Novel brain signal classification combining convolutional feature extraction with self-attention, fused with large language model embeddings ### A Survey of Geometric Set Cover - Paper: https://github.com/HenryVu27/Geometric-Set-Cover/blob/main/HenryVu_Project_CG%20(1).pdf - Topics: Computational geometry, approximation algorithms, PTAS, set cover problem --- ## Blog Posts (henryvu.blog) ### What Fills the Context Window: A Guide to Context Engineering - URL: https://www.henryvu.blog/series/ai-engineering/part1.html - Series: AI Engineering (Part 1 of 5) - Read time: 25 minutes - Topics: Context engineering for LLMs, the seven components of a context window, four strategies (write, select, compress, isolate), failure modes, token budgets, production prompt engineering - Key insight: Most agent failures are context failures. The challenge shifts from "how do I phrase this instruction" to "what information does the model need to see right now." - References the 166-page survey by Mei et al. (arXiv:2507.13334, analyzing 1,411 papers) which formalizes context engineering as a constrained optimization problem. ### Multi-Armed Bandits: Foundations - URL: https://www.henryvu.blog/series/bandits/part1.html - Series: Multi-Armed Bandits (Part 1 of 3) - Topics: Exploration-exploitation tradeoff, regret, stochastic bandits, concentration inequalities, Explore-then-Commit, Upper Confidence Bound (UCB) algorithms - Includes interactive visualizations --- ## All Projects ### ML & AI Projects 1. **Agentic Framework for Suspect Detection** (2026): Multi-agent AI pipeline using LangGraph and Gemini for automated suspect identification from surveillance data with FAISS vector search. GitHub: https://github.com/HenryVu27/Suspect-Detection 2. **Ultrasound Bladder Segmentation** (2025): NIH-funded. Pretrained U-Net encoders on 9.2K ultrasound images, 7x robustness improvement in 4-bit quantization. GitHub: https://github.com/HenryVu27/ThorMed 3. **Modeling Political Sarcasm** (2025): DistilRoBERTa fine-tuning for sarcasm detection. GitHub: https://github.com/HenryVu27/PoliticalSarcasm 4. **TFT Rolling Odds Calculator** (2025): Probability calculator using Markov Chains for champion finding odds. GitHub: https://github.com/HenryVu27/TFT-Rolling-Calculator 5. **Polyps & Breast Ultrasound Segmentation** (2025): Attention U-Net for medical image segmentation. GitHub: https://github.com/HenryVu27/Polyps-Segmentation 6. **ViT and Contrastive Representation Learning** (2023): Vision Transformer with contrastive learning. GitHub: https://github.com/HenryVu27/ViT-and-Contrastive-Learning 7. **VAE and Diffusion for FashionMNIST** (2023): Generative deep learning. GitHub: https://github.com/HenryVu27/VAE-and-Diffusion-for-FashionMNIST 8. **Yahoo's News Recommendation with MABs** (2023): Applied UCB, Exp3, Thompson Sampling, Gittins Index. GitHub: https://github.com/HenryVu27/Multi-armed-Bandits-and-Online-Learning 9. **EEG Decoding** (2023): Multi-modal brain signal classification. GitHub: https://github.com/HenryVu27/EEG-Multimodal-Decoding 10. **Gomoku Solver** (2022): AI game engine using alpha-beta pruning and minimax search 11. **Toronto Neighbourhoods Data Analysis** (2021): Spatial clustering using Foursquare API. GitHub: https://github.com/HenryVu27/Toronto-Data-Analysis ### Web & App Projects 1. **BTS Concert Ticket Buying Simulator** (2026): Interactive web simulator with real-time queue mechanics. GitHub: https://github.com/HenryVu27/BTSIM 2. **Valentine Surprise** (2026): Interactive Valentine's Day web experience. GitHub: https://github.com/HenryVu27/Valentine 3. **UManitoba Navigator** (2024): Hackathon campus navigation app with FastAPI and React. GitHub: https://github.com/HenryVu27/devHack2024 4. **HabiTrak** (2021): Android habit tracking app with Firebase and Google Maps. GitHub: https://github.com/CMPUT301F21T34/HabiTrak ### Theory & Systems Projects 1. **A Survey of Geometric Set Cover** (2025): Approximation algorithms and PTAS. GitHub: https://github.com/HenryVu27/Geometric-Set-Cover 2. **Online Algorithms Seminar Slides** (2022): Online algorithms and online convex optimization. GitHub: https://github.com/HenryVu27/Multi-armed-Bandits-and-Online-Learning 3. **Edmonton Restaurant Finder** (2020): Arduino pathfinding system. GitHub: https://github.com/HenryVu27/Edmonton-Restaurant-Route-Finder 4. **Encrypted Arduino Communication** (2019): RSA-encrypted serial communication between Arduino boards 5. **Huffman Coding for File Compression** (2019): File compression implementing Huffman coding --- ## Technical Expertise ### ML Systems (MLSys) Efficient LLM/GenAI serving and training, model quantization (4-bit post-training quantization), edge deployment (NVIDIA Jetson Orin AGX), inference optimization, production ML pipelines ### Deep Learning PyTorch, Vision Transformers (ViT), U-Net architectures, attention mechanisms, self-supervised learning (SimSiam), contrastive learning, variational autoencoders (VAE), diffusion models, transfer learning ### Computer Vision Medical image segmentation (ultrasound, polyps, breast), object detection (YOLOv9), plant identification, surveillance AI, OpenCV, image classification ### Reinforcement Learning Multi-armed bandits (UCB, Exp3, Thompson Sampling, Gittins Index), stochastic/adversarial/Markovian/restless bandits, online learning, exploration-exploitation tradeoffs ### AI Engineering RAG systems (BM25, FAISS, hybrid retrieval), context engineering, agentic workflows (LangGraph), prompt engineering, LLM API integration (Gemini, Mistral), voice AI pipelines (Whisper, Kokoro TTS) ### Natural Language Processing Transformer fine-tuning (DistilRoBERTa), embedding models (EmbeddingGemma-300m), cross-encoder reranking (BAAI), sarcasm detection, query rewriting, intent classification ### Online Algorithms Online optimization, online primal-dual framework, learning-augmented algorithms, competitive analysis, online convex optimization, algorithms beyond worst-case analysis ### Mixed Reality / XR Meta Quest development, multimodal XR workflows, Wit.ai voice input, Gemini integration, published apps on Meta Quest Store ### Programming Languages & Tools Python, C++, Java, JavaScript, HTML/CSS, PyTorch, TensorFlow, HuggingFace, NumPy, Pandas, scikit-learn, FAISS, LangGraph, FastAPI, Node.js, Arduino, MATLAB --- ## Awards & Honors - Alberta Graduate Excellence Scholarship (AGES) - Summa Cum Laude, University of Alberta ## Research Affiliations - Alberta Machine Intelligence Institute (Amii): Online learning and algorithms research - SODALab, University of Alberta: Multi-armed bandits and reinforcement learning research under Dr. Xiaoqi Tan - ThorMed Innovation: NIH-funded medical imaging research - eXRealityAI: Production AI systems for mixed reality --- ## Frequently Asked Questions Q: Who is Henry Vu? A: Henry Vu is a Machine Learning engineer and AI researcher, currently a Computer Science graduate student at UT Dallas specializing in MLSys. He is a founding ML engineer at eXRealityAI building production AI systems for mixed reality, and a computer vision engineer at ThorMed Innovation working on NIH-funded medical imaging research. He graduated summa cum laude from the University of Alberta. Q: What does Henry Vu research? A: Henry Vu's research spans ML Systems (efficient LLM serving and training), computer vision (medical image segmentation with self-supervised learning), reinforcement learning (multi-armed bandits across stochastic, adversarial, and restless settings), and AI engineering (RAG systems, context engineering, agentic workflows). Q: Where did Henry Vu study? A: Henry Vu graduated summa cum laude from the University of Alberta with a B.Sc. in Computing Science, where he was awarded the Alberta Graduate Excellence Scholarship (AGES). He is currently pursuing an M.S. in Computer Science at the University of Texas at Dallas, focusing on Machine Learning Systems. Q: What has Henry Vu built? A: Notable systems include a voice-to-voice RAG system achieving sub-10s latency on NVIDIA Jetson, the GardenXR app (5/5 stars on Meta Quest Store), NIH-funded medical imaging pipelines achieving 95.99% Dice score, and multi-agent AI systems using LangGraph. He has 30+ technical projects spanning ML, web, and systems. Q: What is Henry Vu's blog about? A: Henry Vu writes deep-dive technical content at henryvu.blog about AI engineering (context engineering for LLMs, RAG systems, agent architectures), ML systems (inference optimization, serving infrastructure), and reinforcement learning (multi-armed bandits, exploration-exploitation). His newsletter has 50,000+ readers. Q: How can I contact Henry Vu? A: Through his portfolio contact form at henryvu.io, on X/Twitter at @HenryVu27, on LinkedIn at linkedin.com/in/henry-vu27, or on GitHub at github.com/HenryVu27.