David Hu-Liu
About Me
01Hi, my name is David Hu-Liu, a Computer Science Student @ UWaterloo and Software Engineer. I have a strong interest in Machine Learning/Data Science, Mathematical Finance, Full-Stack Development, Graphic/UX design, and creating products that address real-life problems.
As of right now, I am currently most interested in Software Development opportunities in the Fin-Tech space; Combining my Investment interest alongside techniques learnt in SWE. And in in my long-term vision, I hope to try Sales Solution Architecting, which may pose as quite the hustle.
I wish to continue learning because I believe technology, including software, has the capacity to shape our future and address societal problems such as sustainability & equality/access. Technology has historically been a catalyst in improving our quality of life, and the future will be no different.
With open arms, I seek to collaborate with others with the shared mindset that technology can drive positive social change. Outside of my career goals, I am quite passionate about fitness and hybrid athletics.
EDUCATION AND CERTIFICATIONS
Bachelor of Computer Science
See More Detail Here
[ 2023 - 2028 ]
Harvard Pre-College Program: Mathematical Elements of Data Science and Artificial Intelligence
Experience
02Researcher / Architect | Power Grid
[Mar 2025 - Present]
• Developing an AI-powered power grid intelligence tool to optimize data center energy site selection, leveraging Pandas, NumPy, SciPy
for large-scale energy datasets (EIA, FERC, ISOs) and TensorFlow/PyTorch for predictive modeling.
• Conducting Quantitative PE Real Estate Research with 2Sigma/JW Capital Management VP, identifying
Multi-Billion Dollar Investment inefficiencies in energy infrastructure.
• Designing geospatial analysis pipelines using Leaflet.js, Mapbox to visualize real-time power availability, enhancing
decision-making for investors in energy-intensive real assets
AI / Software Engineer | CosmicBase
[Jan 2025 - March 2025]
• Building and Training a Platform to enable fractional ownership of Luxury Properties across NA(Incomplete)
• Superbase, MERN, Full-Stack , etc.
Software Engineer | Tyce.io
[Dec 2024 - Mar 2025]
• Architecting a scalable full-stack document generation platform using modular microservices, reducing API latency by 40% and enabling seamless integration of AI workflows for 10,000+ sales documents potentially.
• Engineering a RAG pipeline integrating proprietary sales data with LLMs (GPT-4, Claude), achieving 92% accuracy in context-aware contract generation and reducing manual editing time by 15 hours/week per sales team.
• Researching Python-based AI agents using LangChain and FastAPI to automate pricing prediction workflows, analyzing 500+ historical deals to generate quotes with 88% accuracy compared to human benchmarks.
Co-Founder & CEO | Flair Fashion Studio
[July 2024 - Present]
• Spearheaded AI-driven fashion analytics platform, securing $25,000 USD Microsoft Incubator Grant and achieving 80% MoM user growth by aligning product roadmap with gaps identified in 300+ fashion studio brands (D&G, VW, Acne Studio, Saint Alexander).
• Architecting proprietary trend-prediction algorithm using PyTorch, reducing design-to-market latency by 40% for partner brands through real-time analysis of 10M+ social media/runway data points.
• Training multimodal LLM (text + image) on 10TB of studio brand archives and consumer sentiment data, achieving 94% accuracy in predicting next-season color/fabric trends validated against Vogue Business benchmarks.
• Slashed customer acquisition costs (CAC) by 55% via hyperlocal sentiment analysis of Instagram/TikTok trends, enabling targeted influencer campaigns with 18% conversion rates.
Software Developer/Prompt Engineer | DataAnnotations
[Aug 2024 - Oct 2024]
• Engineered and Trained LLM’s to produce higher quality code across 5000+ lines per day, distributing annotations to
10k+ developers for review, through Achilles Pillars Response Analysis and prompt engineering.
• Writing and testing software apps and features using python libraries like Pytorch, Numpy, Tensorflow, react
frameworks like Next.js and Bootstrap, Restful API’s, and databases like SQL, MongoDB.
• Training LLM’s on their ability to complete Leetcode/Hackerrank style questions in Java, Python, C, and C++.
Full-Stack Engineer | Aview International
[May-Sept 2024]
• Developed and Designed a comprehensive AI video editor tool that 300+ famous content creators leverage for
customized videos (6+ Integrated pages, 10+ Additional features)
• Restructuring main Full-stack app and Admin app, through addressing 30+ major bugs, 15 new web pages, 35+
component additions using Next.js, Nest.js, Mongoose, Redis, and RestfulAPI’s with Postman
• Proposed and initiated integrating lip sync and deep-fake ML, handling resource hosting infrastructure, cloud and
database management using Pytorch, Keras, matplotlib
Sales Engineer | Aview International
[May-Sept 2024]
• Led the initiative to take on multiple clientele, landing 3 major content creators at Collision2024 networking conference;
Loren Gray, social media superstar/singer(80MIL+ Followers). Hafu Gao(8MIL+ on YT). SupaHotFire(500k+ on YT)
• Partnering with client outsourcing companies like SAAS Founders to discuss automating of client attraction process, with
potential of 30% increase in clientele, using search engine outreaching algorithms.
Amateur Poker Player | UWPC, VCLUB
[Sept 2023 - Present]
• Achieved #1 Ranking (top 0.001%) in Winter 2024 Tournament Series [University of Waterloo Poker Club], outperforming 500+ competitors with a 28% ROI by applying advanced equity modeling and risk-adjusted decision frameworks.
• Studying dynamic capital allocation strategies across 50+ high-pressure tournaments, optimizing bankroll management to limit drawdowns to 5% of total liquidity while maintaining 99th percentile profitability.
• Advised C-suite Executives at Founders of Toronto Poker Club & VClub on portfolio optimization tactics, translating poker-derived risk/reward frameworks into venture capital deal structures with projected 15% alpha generation.
Data Science Student | Havard Pre-College [July 2023]
• Proposed a Grayscales strategy, yielding 98% accuracy by epoch 10 with training data for CIFAR100 dataset
• Leveraged Principles of CNN(loss function, activation function, convolution layers, pooling, etc) to train a deep neural
network classification model on 60,000+ data points using Tensorflow, Numpy, and Pytorch
• Spearheaded a team of 4, to achieve the best neural network in our Harvard Pre-College class
Retail/Equity Trader | Interactive Brokers [Jan 2021 - Present]
• Generated 20% annualized returns (vs. SPY 14%) via systematic covered call strategies, leveraging dynamic delta-hedging to mitigate tail risk during 2022–2023 volatility regimes.
• Deployed volatility arbitrage strategies (skew trades, dispersion) across 50+ underlyings, exploiting mispricings in IV vs. realized vol to capture 15% alpha annually.
• Automated Greeks-based hedging (delta/gamma/vanna) via Python scripting, reducing manual rebalancing labor by 80% while maintaining 99% hedge effectiveness.
• Backtested Monte Carlo simulations (10K+ iterations) to optimize multi-leg spreads, improving risk-adjusted returns by 22% through dynamic strike selection and expiry laddering.
Projects
03
StoryNook App
Coming Soon... Hold On... Let Me Cook
PROJECT KLEAN
Klean is a digital waste cleaning web app that offers Code Optimization along with Data Analysis for emissions. By analyzing and optimizing Python code to improve its efficiency and sustainability, it aims to reduce excess computational movement at an enterprise level to save energy. Klean identifies memory-intensive operations, redundant computations, and suboptimal data structures, and suggests improvements.
•Engineered an enterprise-level digital waste, code cleaning solution achieving 60% reduction in computation waste, 45%
decrease in carbon emissions across projects hosting on AWS/Google Cloud
• Leverages AI refactoring and Microsoft CodeBERT, optimizing over 10 million lines of code at maximum capacity for
firms and institutions
HANDPILOT
In a world where communication with technology is essential, and some are less fortunate, we proposed Handpilot: Allowing Users to Navigate the computer effortlessly without the need for a mouse. Let your hands glides freely without the need to apply strain to potentially injured or painful joints.
• Led a team of 3 to pitch a hand-controlled, motion detected virtual mouse, addressing Carpal Tunnel, Fractures, Tendonitis,
etc, with a 95% success rate in improved patient autonomy
• Implemented Python Libraries like OpenCV, Mediapipe, Pyinput, for camera detection and mouse inputs to increase
user productivity by 80%
VERBATE
Become Interview-ready easily with Verbate. Speaking itself is a skill beyond just what you know. Practice your public speaking skills using this chatbot that allows for voice recognition. The bot will give you effective interview questions based on your job requirments and personalized prompts. It will then give you a final evaluation of how well you did on your practice interview, and how likely you are to do well in the future.
• Utilized Node.js, Next.js, React, JS, AWS poly/GPT-4 API to create an interview-style voice interactive chatbot
that achieves 98% reported improved interview confidence, and supports 25+ languages
• Designed a custom prompting system which improved interview retention time by 40% and reduces prompting hassle by
1100 characters compared to plain GPT-4
• Restructured server hosting to increase response loading speed by 30%
Image Recognition Machine Learning AI [CIFAR 10]
As our final assignment in my Harvard Pre-College program, we were instructed to traing our own CNN model and compete for accuracy in small groups.
Leveraging what we learned about topology, calculus, 3D geometry, and elementary statistics our group achieved the highest accuracy (98%) in the class, by approaching the problem using a simplistic activation layer set up and our clever grayscale coversion strategy that I initially proposed. The image conversion to grayscale would reduce a pixel's vector values from the range of 0-255 to just 0-1, allow for significantly easier processing after each additional convolution layer.
My Three.js Personal Portfolio
The website you are currently reading, utilizes blender for the room model, three.js for rendering and camera movements, GSAP for high quality animations, and css/html.
I had originally tried to create an interactive room using Spline 3d. I gave up, and found this three.js room model I liked instead.
Chatbot NLTK Machine Learning AI
Utilizing Python's Natural Language Took Kit library and SGD optimization to create a silly language processing bot that randomizes its responses based on training data.
For example, entering "hi, how are you", will be processed and identified as a "greeting" intention and will output any of the predetermined "greeting" responses I set. In this regard, the bot is limited to the types of intentions pretedermined by me. The maximum training accuracy I achieved was 91%. I left the overall project unfinished and faulty after becoming discouraged realizing its harsh limitations.
SKINSIGHT
Utilizing flask, OpenCV for image processing, pytorch and tensorflow, and html/css for front-end to create a web app that predicts the likelihood of the different types of skin cancer.
This project was inspired by the excruciating wait-time of the Canadian Healthcare System. By hopefully providing the opportunity for self-diagnosis, it could speed up healthcare and also contribute to cancer mitigation in the future.
WhatsInThis?
A modern problem that persists is not knowing the nutrition of meals when going out to eat.
Website utilizes React.js, Tailwind CSS, and the Nutrition information API from NutritionIX's dataset to create a simple website that fetches for the nutrition value of nearly any food or meal. Back-end login authenitcation and meal-saving feature are still a work-in-progress.
The SIMON Game
Simon is a childrens Pattern Recognizing Memory Game. The objective is to recite the pattern, as flashed by the different colors. This project was a good opportunity to practice basic logic in js using the jquery library. A sequence is randomly generated and added to a list, and continuously checked if the users input matches it. Once the user cannot match the next sequence, the player loses.
CONTACT
04If you'd like to get in touch, please feel free to connect with me on LinkedIn. On LinkedIn, I am actively trying to engage in coversation about stocks/trading algos, software development, and finance in general. I am available for interviews and I look forward to connecting with other professionals in the computer science field.
Github
I can be found on Github, where all of my projects are, and where I am trying to actively grow. I try stay up to date with the newest technology, and continue learning whenever I can!
Devpost
I try to compete in Hackathons whenever I can. You can find any new hackathon submissions here.
