career_quest.exe
MAIN QUEST
JUN 2026 – NOW
Exploring the Code World 😋
the open internet · self-directed
Building more fun things — starting with the OS you are standing in: a live-streaming slime, 72 evolving pikmin, a nightmare boss, a smartwatch edition, and a terminal with a door in it. Zero frameworks, all heart.
🖥️ this entire OS, hand-rolled
⌚ wrist edition + cloud sync
🐇 terminal door + mini-CTF
CO-OP QUEST
MAY 2026 – JUN 2026
Open-Source AI Agent Engineering Lead
RustChain Ecosystem · Independent Contributor
Ran a self-improving fleet of coding agents that discovered issues, shipped PRs, answered reviewers, and learned from every merge — with a human hand on all the money-path and safety levers.
🏆 merged security & wallet PRs
🔁 self-improving AgentOps loop
🛡️ rate limiting + payout safety
▸ full quest log
- Designed and operated an autonomous AI-agent engineering workflow for RustChain open-source development, coordinating specialized agents for issue discovery, PR implementation, review monitoring, maintenance updates, regression testing, and PR communication.
- Built a self-improving AgentOps loop that automatically monitored reviewer comments, CI failures, maintainer feedback, bot labels, and PR state changes; summarized recurring review patterns; updated internal task-selection and implementation rules; and reused those lessons to improve future PR scoping, validation, and review responses.
- Implemented an automated maintenance-response pipeline: when reviewers requested changes, the Druid agent detected the requested action, analyzed the relevant code path, generated targeted code/test updates, ran validation checks, and posted follow-up maintenance comments summarizing the fix and test evidence.
- Developed a repeatable PoC-to-fix delivery system covering issue selection, root-cause analysis, failing-test generation, minimal patch design, local validation, CI-aware debugging, PR body drafting, and reviewer-response preparation.
- Delivered merged RustChain contributions across security, wallet/reward integrity, infrastructure reliability, and API usability, including:
- Rate limiting for the core RPC/API handler with per-IP isolation, trusted API-key handling, CORS preflight support, and regression coverage.
- Event faucet claim-code flows with admin-gated code creation, one-time redemption, transfer status tracking, idempotency handling, and retry-safe failure behavior.
- Beacon Atlas public profile routing that preserved admin-only fields while exposing redacted public agent profiles and preventing alias shadowing.
- Validator and explorer metrics improvements, including attestation effectiveness reporting, block filtering, and reorg-resistance metrics.
- Documentation and installation-flow fixes that improved contributor onboarding and reduced broken-link or stale-path friction.
- Focused heavily on money-path and security-sensitive surfaces, including wallet balances, pending confirmations, settlement correctness, bridge/wRTC flows, faucet payout safety, reward accounting, idempotency, concurrency, and admin-boundary protection.
- Designed Druid/Chair operating rules for high-signal RustChain work: prioritize financial/security/reward impact, avoid speculative rewrites, prevent duplicate PR coverage, preserve payout and production-wallet safety, and keep each PR small enough for efficient maintainer review.
- Built automated learning summaries from merged PRs, reviewer feedback, failed attempts, and maintainer preferences, allowing the agent workflow to continuously refine issue targeting, claim strategy, implementation boundaries, testing standards, and PR communication quality.
- Used AI agents as autonomous engineering accelerators while maintaining human governance over safety boundaries, production-impact decisions, reward/payout logic, and final PR strategy.
MAIN QUEST
JUN 2025 – MAR 2026
Systems/LMS/Full-Stack Programming Lead
ARC in Action · Courage to Act Foundation & Sea Slug Digital · Canada
Solo-built a production Moodle LMS on AWS that runs itself: auto-heal, backups, disaster recovery — designed so a fully non-technical team never has to touch a server.
☁️ AWS EC2 + S3 + Nginx + Redis
♿ WCAG 2.1 AA verified
🩹 auto-heal + nightly DR
↗ visit the live production site
▸ full quest log
- Built and launched a production Moodle on AWS EC2 with Nginx/PHP-FPM/MariaDB/Redis, Let’s Encrypt + HSTS, and environment hardening; configured CloudWatch logs/alarms and OS-level health checks for reliable ops.
- Implemented multi-layer security: firewall hardening (UFW/Fail2ban/SSH lockdown), least-privilege IAM, and a locked-down plugin installation workflow with integrity checks and role-scoped access to reduce attack surface.
- Designed a hands-off maintenance pipeline for a non-technical team: cron-scheduled automatic disk housekeeping; daily web-performance health checks with auto-remediation; Moodle cron health monitoring; database & course content backups to AWS S3 (versioning/lifecycle + DR replication); nightly instance snapshots/auto-clone; auto-heal on service failure; and email alerts when manual intervention is required.
- Set up SMTP and end-to-end login/upload flows; added OPcache and Redis caching to stabilize performance under load.
- Led UI/UX & accessibility hardening to align with WCAG 2.1 AA: visible "Skip to main content", correct landmarks/heading structure, ARIA-compliant menubar (Tab to enter/exit; Left/Right to traverse), standardized :focus-visible styling, and AA-safe palette.
- Verified accessibility with WAVE and Accessibility Insights for Web (FastPass/Tab Stops) across Home, Dashboard, Course main, and Page views; resolved color-contrast and keyboard-navigation blockers.
- Documented an authoring guide (images alt text, accessible names for icon-only controls, and the MP4 + .vtt workflow) and created copy-paste snippets to reduce content-level defects.
CO-OP QUEST
FEB – MAY 2024
Data Engineer
HyperGAI · Singapore · Internship
Ran the data side of a multimodal GenAI lab: 200+ annotators across three languages, an S3 data warehouse, and video pipelines that measurably raised dataset quality.
👥 200+ annotators, EN/中文/粤語
📈 +26% resolution · +17% length
🗄️ S3 data warehouse
▸ full quest log
- Directed cross-regional coordination with third-party annotation teams (200+ annotators based in Hong Kong) for high-volume video labeling projects; maintained multilingual communication in English, Mandarin, and Cantonese to ensure real-time clarity.
- Orchestrated large-scale annotation schedules under tight timelines, managing project milestones and deliverables across multiple stakeholders.
- Implemented Google Sheets API integrations to establish a real-time collaborative environment, enabling instant multi-user data updates and improving team agility.
- Led productionization of data pipelines supporting AIGC model development; optimized video filtering workflows, boosting average output resolution by 26% and length by 17%.
- Built and maintained AWS S3-based data warehouse, developing efficient ingestion, storage, and retrieval processes for large-scale multimodal datasets.
- Designed Nginx-based microservices to address annotation platform performance bottlenecks, improving system loading speed and streaming smoothness.
- Supported real estate chatbot development by testing and deploying deep learning models into production, collaborating cross-functionally to improve system dialogues based on customer feedback.
- Developed quality assurance pipelines for AIGC datasets, applying Optical Character Recognition (OCR) and optical flow analysis to enhance input quality for downstream ML systems.
RESEARCH DUNGEON
JAN – APR 2024
Financial Time Series Analyst (R)
Nanyang Technological University
Hybrid ARIMA-GARCH/SARIMA forecasting stack with walk-forward validation — grid search did the grinding so humans didn't have to.
🎯 12-step RMSE < 2.8
⏱️ −70% tuning time
📊 +18% vs baseline
▸ full quest log
- Designed and automated a hybrid ARIMA-GARCH-SARIMA forecasting pipeline with full grid-search hyperparameter tuning, streamlining model deployment processes and reducing manual optimization time by 70%.
- Developed modularized ETL pipelines using R’s forecast, rugarch, caret, and zoo libraries to process large-scale financial time series datasets in both structured SQL and semi-structured NoSQL storage formats.
- Applied walk-forward validation techniques to productionize time-series models with real-world deployability, achieving 12-step prediction RMSE < 2.8.
- Engineered temporal feature extraction workflows, integrating database timestamp metadata into training pipelines, thus enhancing predictive model accuracy by 18% over baseline models.
RESEARCH DUNGEON
AUG 2023 – APR 2024
Graduate Research Assistant — Real-time Defect Detection
NTU Real-time Industrial Part Monitoring Project
Real-time YOLOv8 defect detection for manufacturing QA — benchmarked against YOLOv5 and Faster R-CNN, wired into the actual factory monitoring loop.
👁️ >82% mAP50-95
🏭 +18% throughput
☁️ S3 defect analytics
▸ full quest log
- Designed and productionized a real-time defect detection system using YOLOv8n, achieving >82% mAP50-95 across multiple industrial classes, ensuring high scalability and low latency for manufacturing QA applications.
- Fine-tuned and benchmarked object detection models (YOLOv5, YOLOv8, Faster R-CNN), iteratively optimizing for precision and inference speed, simulating ETL-like model retraining pipelines for continuous data ingestion and model updates.
- Collaborated with cross-functional engineering teams under Agile methodology to integrate the defect detection models into the automated industrial monitoring systems, improving operational throughput by 18%.
- Implemented cloud storage integration for model outputs, supporting traceable defect analytics and scalable cloud-based reporting solutions via AWS S3 buckets.
RESEARCH DUNGEON
JAN – APR 2023
Undergraduate Research Assistant — Coyote Mange Detection
University of Alberta
Almost no labels? Semi-supervised ResNet-18 more than doubled recall on wildlife camera images.
🐺 recall 25% → 52%
🧪 semi-supervised pipeline
▸ full quest log
- Pioneered semi-supervised learning pipelines leveraging ResNet-18, enhancing scarce data utilization and improving model recall from 25% to 52%, reflecting production-level deployment considerations under limited labeling scenarios.
- Automated the preprocessing and augmentation pipelines by building custom Python modules, including background removal, object detection re-centering, and contrast normalization.
- Implemented model evaluation APIs to streamline iterative experiments and standardize hyperparameter tuning across training runs, aligning with Agile DevOps practices.
- Designed initial database schemas (Postgres) to organize annotated datasets, optimizing storage retrieval speed for high-resolution images.
RESEARCH DUNGEON
SEP 2022 – MAY 2023
Undergraduate Research Assistant — EEG-to-Image Modeling
UofA Human-Robot Visual-Inertial Monitoring Project
Turned raw EEG brainwaves into images a CNN could read — then served the models two different ways just to prove the pipeline was framework-agnostic.
🧠 >70% trend prediction
⚡ TF Serving + PyTorch APIs
▸ full quest log
- Developed an end-to-end pipeline converting EEG signal streams into images for deep learning analysis, using TensorFlow convolutional autoencoders to achieve >70% neurological trend prediction accuracy.
- Built microservice-style model deployment architectures utilizing both TensorFlow Serving and lightweight PyTorch APIs, demonstrating adaptability to multiple ML deployment frameworks.
- Integrated time-synchronized visual-inertial data streams using custom Python APIs, emulating real-time data ingestion and transformation tasks similar to production data pipelines.
- Authored a technical handbook standardizing data preprocessing and signal transformation, contributing to infrastructure documentation practices akin to DevOps-quality standards.
inventory.sav
CHARACTER STATUS
CODE92
AI / ML90
CLOUD / SRE88
DATA ENG86
UI/UX + A11Y85
COMMS ×3 LANG95
EQUIPPED ITEMS
Python●●●●●
AWS S3/EC2●●●●●
Agent Ops●●●●●
PyTorch / TF●●●●○
Nginx●●●●○
SQL / NoSQL●●●●○
Security●●●●○
WCAG A11y●●●●○
ITEM BOX
Pandas
NumPy
HuggingFace
C / C++
R
Azure
MySQL
MongoDB
Redis
Elasticsearch
Snowflake
YOLO
Faster R-CNN
ARIMA-GARCH
Jenkins
GitHub Actions
CloudWatch
PHP-FPM
MariaDB
OPcache
TLS / HSTS
UFW / Fail2ban
Bash / cron
WAVE
A11y Insights
LaTeX
REST APIs
Agile
VOICE LINES
🇬🇧 English · fluent
🀄 Mandarin · native
🥟 Cantonese · native
🥐 French · studying
stream_chat.log
You're watching yongshan_dev — cute UIs on serious infra. Type !hire to copy her email ♡
ask_me.chat
terminal.exe
yongle_search.exe
slime_run.exe
score 0
★ best 0
SPACE / tap to jump · avoid the bugs, catch nothing, just survive
hall_of_slime.exe
…
🏆 top-10 run (0)! sign it:
LOCAL TOP 10
🏅 ACHIEVEMENTS
💾 CLOUD SAVE
photo_album.exe
wrist_link.exe
dreamlog.exe
pikdex.exe
🎨 HUE_WHEEL.SYS
0%
interview_scheduler.exe
The HR fairy sent you! Pick a slot in Yongshan's real availability below — once you confirm, both of you automatically receive a confirmation email. ♡
slime_live.exe
gift shop ♡ tap = send
camera runs 100% locally — it spots waves, nods, peek-a-boo, guests arriving… and even compliments your drink by its colour. nothing is recorded or uploaded, ever.
education_awards.txt
🎓 EDUCATION
MSc in Artificial Intelligence
Aug 2023 – Apr 2024
GPA 4.21 / 5.0
BSc in Computer Science (Minor: Psychology)
Sep 2019 – Apr 2023
GPA 3.30 / 4.0
🏆 ACHIEVEMENTS UNLOCKED
- ★ International Student Scholarship — $2,000/year, 2019–2023, University of Alberta
- ★ Graduate with Distinction — Top 5% of graduating class, University of Alberta (2023)
start_here.txt
Welcome to the desktop! Things to try:
- ♡ pet the slime — click combos raise your fan count
- 🍬 feed it, ⭐ play with it, 🌙 let it nap
- 🖐️ drag the slime around · double-click for a trick
- 💗 ask_me.chat — an offline bot that answers questions about me
- 🖥️ terminal.exe — type
help… orneofetch - 🔍 type anything in the address bar — this browser searches!
- 🌙 in dark mode the slime sometimes goes to bed — knock to wake it
- ⚠️ petting it 30+ times may cause a love overflow…
- 🔴 the red dot? press it. trust me.