Henry Zhang

Co-founder, CTO & CPO

• Co-founded AI exam-prep platform; indexed 50K+ questions from 500+ PDFs with 92%+ positive beta ratings. Full-stack owner across Django/DRF backend, Nuxt 3 frontend, and infra (Docker Compose, Redis, Postgres, Qdrant).
• Engineered 4-stage LLM+OCR parsing pipeline—text extraction, OCR region detection, semantic matching, automated crop/merge—to convert image-heavy exam papers into structured question-level data at scale.
• Implemented hybrid search combining Postgres trigram indexes and Qdrant vector similarity with syllabus-aware query expansion; added photo-search (OCR-to-embedding) enabling 'search by screenshot.'
• Built AI chat interface on Convex with LLM tool-calling (search, mark-scheme fetch, practice-set generation), streaming responses via Redis streams + Channels websockets, and per-user credit quotas with OpenRouter orchestration.
• Shipped interactive practice-set builder with Fabric.js canvas, image cropping/editing, auto-layout, undo history, and one-click PDF export of questions paired with mark schemes.
• Integrated auth and billing end-to-end: Ory Kratos identity, RBAC, Stripe subscriptions with credit metering, invite-gated rollout, and rate limiting.
• Tech: Django, Nuxt 3, Convex, Pinia, Qdrant, PostgreSQL, Redis, Fabric.js, Ory Kratos, Stripe

ML for Quantitative Trading: STAR Market Prediction

• Built Random Forest model achieving 195% cumulative return and 89.9% annualized return over 6-year backtesting period (2019–2025).
• Outperformed STAR 50 index by 0.54% daily alpha using 8 engineered features from 556,000+ datapoints across 589 stocks.
• Designed rigorous evaluation: chronological train/test split, feature importance analysis, comparison with XGBoost baseline.
• Managed risk with 31.8% maximum drawdown; documented methodology in technical report.
• Tech: Python, pandas, scikit-learn, XGBoost

Kaggle Machine Learning Competitions

• MITSUI&CO. Commodity Prediction Challenge (Bronze, 151/1,711): Built time-series ensemble (LightGBM/XGBoost/CatBoost) with lag features and per-target Ridge blending to predict 424 commodity-spread targets.
• Hull Tactical Market Prediction (Ongoing): Featured competition modeling stock returns under volatility constraints.
• Playground Series placements: Loan Payback Prediction (Top 9%), Diabetes Prediction Challenge S5E12 (Top 12%), and Predicting Student Test Scores (Top 12%).

Team Captain & Software Lead

• Developed real-time projectile ballistics solver using iterative numerical methods to compute launch angle, flywheel velocity, and turret lead for moving targets.
• Implemented moving-shot compensation with kinematic decomposition of robot velocity to adjust hood angle and flywheel speed.
• Built Limelight AprilTag + OpenCV solvePnP 3D pose estimation with distance-based aiming offsets and localization fallback.
• Created autonomous stack with state-machine routines, Bézier trajectories, and alliance-agnostic field mirroring.
• Engineered command-based architecture with optimized scheduler, three-dead-wheel odometry, and pure pursuit path following.
• 2024–25 Season: Innovate Award #1 at FTC World Robot Contest China Off-Season Championships; 2025–26 Season: 1st Place Qualifiers, Finalist Alliance Captain, Inspire Award #2; 3rd Global TeleOp, 19th Global Overall.
• Tech: Java, FTC SDK, OpenCV, Limelight, Pedro Pathing

AI for Robotics: Automated Sorting System

• Engineered autonomous recycling classification system combining YOLO11 computer vision and robotic control in PyBullet 3D physics simulation.
• Built synthetic data pipeline generating thousands of labeled training images from zero real data — solved cold-start problem.
• Implemented inverse kinematics and FSM control for KUKA robotic arm to grasp and sort objects from moving conveyor belt.
• Developed camera calibration pipeline transforming 2D pixel coordinates to 3D world coordinates for accurate trajectory prediction.
• Tech: Python, OpenCV, PyBullet, YOLO, NumPy

Contributor & Member — maple-sim Open-Source

• Contributed to Maple-Sim (103 stars, 56 forks), an open-source dyn4j-based physics engine for FRC robot simulation; recognized by teams as 'crucial to help all of FRC.'
• Designed drivetrain and swerve simulation with wheel friction, centripetal forces, and motor controller modeling at 200+ Hz.
• Built a physics-based 'shooting on the fly' system for an FRC swerve robot, combining odometry/vision pose, velocity compensation, and real-time heading control to score while moving.
• Implemented multi-camera AprilTag pose estimation with weighted filtering, ambiguity thresholds, and custom MLE pose fusion.
• 2025: Innovation in Control Award (Canadian Pacific), Engineering Inspiration Award (Shanghai), World Championship qualifier (Archimedes Division).
• Tech: Java, WPILib, dyn4j, PathPlanner

AI-Powered Air Hockey Robot

• Built real-time visuomotor control system running ACT (Action Chunking Transformer) imitation learning policy at ~30 FPS.
• Integrated OpenCV camera stream with SO-101 robotic arm via LeRobot controller.
• Implemented CVAE-based trajectory prediction with transformer encoder-decoder architecture fusing robot state and image embeddings.

Open-Source Contributor

• Contributed to 4 Obsidian plugins (AI workflows, Jupyter integration) and Dayflow (5.8k stars)—overhauled settings UI and fixed 5 bugs in SwiftUI.

Full Stack Game Programmer

• Co-founded and scaled game franchise to 250,000+ plays; won Best Game Design Award at Roblox Nationals (2021).
• Engineered backend data persistence and gameplay systems in Lua; iterated on player feedback to drive retention.
• Tech: Lua, Roblox Studio, Git/GitHub

Co-founder, CTO & CPO

• Led 3-person dev team through full product cycle: requirements, architecture, sprint planning, code review, launch, and iteration.
• Shipped invite-gated beta to 40–50 users; achieved 92%+ positive ratings and 50K+ questions indexed across IGCSE/A-Level subjects.
• Defined product roadmap including difficulty ranking, teacher dashboards, mobile revision mode, and AP/IB/SAT expansion.

Team Captain & Outreach Coordinator

• Grew team from 7 to 15 members as Captain (114% growth).
• Improved funding by 1,245% via 14 sponsors; achieved 80% financial independence through structured sponsorship pipeline.
• Expanded outreach events by 550%, including partnership with Tencent Volunteering Teaching to deliver online STEM courses.
• Coordinated exchanges with 9 FTC and 4 FRC teams nationwide; led #FIRSTLikeAGirl initiative to promote female participation in STEM.
• Directly impacted 900+ individuals through STEM teaching; accumulated 192+ team volunteer hours across 4 schools.
• Demonstrated at Maker Faire Shenzhen (40,000+ attendees); distributed 500+ educational STEM kits.

Student Mentor (Software)

• Mentored software development through code reviews, debugging support, and iterative control/auto tuning.
• 2025–26 Season: Control Award Winner at China Championship; 1st place qualifier (Beijing #1) and Finalist Alliance Captain.

High School Diploma

• GPA: 3.97 / 4.0

Honors & Awards

• Kaggle: Mitsui & Co. Commodity Prediction Challenge — Bronze Medal (151/1,711 teams); Hull Tactical Market Prediction (Ongoing)
• HiMCM: Meritorious Award (Top 12%) — Environmental Impact of Sport Mega-Events (2025)
• FRC (2025): Innovation in Control Award (Canadian Pacific); Engineering Inspiration Award (Shanghai); World Championship Archimedes Division
• FIRST Tech Challenge (2025–26): Inspire Award 2nd Place, 1st Place Qualifiers (6-0), Finalist Alliance Captain
• FIRST Tech Challenge (2024–25): Innovate Award 1st Place at World Robot Contest China Off-Season Championships
• USACO Platinum Division, Certified (2026)
• Best Game Design Award, Roblox Nationals High School Division (2021)