约 10 分钟Quant finance
TradingAgents：多智能体交易框架的决策稳定性与投资表现
我用月度回测和重复运行，检查 TradingAgents 的收益取舍、评级一致性，以及它是在下跌前识别风险，还是下跌后才提高防御。
阅读文章
Sam Zhong
钟佳有/ Zhōng Jiā yǒu /
Harnessing the power of AI to revolutionize software and finance.
Software Engineer · FinTech · NLP Research
📍TorontoUWaterloo CSUofT Master of Math Finance
Skills
Technical skills across the stack.
From LLM research to full-stack products — here's what I work with.
Languages
PythonC++CTypeScriptJavaScriptKotlinReact NativeSQL
AI / ML
PyTorchScikit-learnLangChainFine-tuningHugging Face
Web & APIs
ReactNext.jsNode.jsFastAPIRESTGraphQLTailwind CSS
Cloud & Tools
GitDockerKubernetesAWSLinuxCI/CDPostgreSQLMongoDB
Certifications
AWS Certified Developer – Associate
Amazon Web Services
Issued Nov 2025
AWS Certified AI Practitioner
Amazon Web Services
Issued Oct 2025
AWS Certified Cloud Practitioner
Amazon Web Services
Issued Aug 2025
Experience
Where I've worked and what I've built.
Internships, research, leadership, and the occasional close call.
BMO Financial Group
UpcomingIncoming Quantitative Developer
Jan 2027 – Apr 2027
Toronto, ON
Stealth Startup
StartupFounder
Jul 2026 – Present
China
- Built a cross-market alpha platform integrating U.S. and Chinese sector data for 100+ retail investors.
- Developed an Alpha158-style XGBoost sector-ranking model that achieved 12.4% annualized return, a 1.18 Sharpe ratio, and -9.6% maximum drawdown in backtests.
Alpha ResearchPythonNext.jsAlibaba Cloud
The Hong Kong University of Science and Technology
InternshipSoftware Engineer
May 2026 – Aug 2026
Hong Kong, Hong Kong SAR · Remote
- Built a LangGraph multi-agent pipeline for market analysis, cutting end-to-end analysis time by 85%.
- Reduced median API latency by 35% across 70+ FastAPI endpoints through Redis caching.
- Persisted agent state and trade logs in Supabase Postgres, with Sentry alerts configured to report issues within 5 minutes.
PythonLangGraphFastAPIPostgreSQLRedis
Health Canada
InternshipSoftware Engineer Intern
May 2025 – Apr 2026
Ottawa, ON
- Rebuilt a Streamlit MVP with TypeScript, FastAPI, Redis caching, and load-balanced Azure Container Apps replicas, supporting growth from 5 to 600+ daily users.
- Scaled vLLM on Kubernetes with GPU requests and tensor parallelism, achieving 6× inference throughput.
- Tracked request latency, errors, and Kubernetes workloads in Azure Monitor for 24/7 production monitoring.
TypeScriptFastAPIPostgreSQLKubernetesvLLMAzure
Health Canada
InternshipMachine Learning Engineer
Jan 2025 – Apr 2025
Ottawa, ON
- Replaced fixed-length chunking with hierarchical, tokenizer-aware splitting and metadata-enriched serialization, improving retrieval accuracy by 84%.
- Built a hybrid-search RAG system with MilvusDB and reranking, raising answer precision from 52% to 89%.
- Evaluated retrieval quality, faithfulness, and answer relevancy with an LLM-as-a-Judge pipeline in under an hour.
- Configured LangSmith tracing and 5-minute Slack alerts for RAG error rates, latency, and cost spikes.
PythonAzureLangChainHugging FaceVector DBStreamlitDocling
Sep 2024 – May 2026
Sydney, NSW
- First-author EMNLP 2025 paper proposing EthosAgents — a role-playing persona framework for pluralistic alignment in healthcare LLMs, improving value diversity by 40% over prior approaches on the VITAL benchmark.
- First-author paper on latent value alignment via internal activation steering with controlled value vectors, achieving 30% gains over persona-based methods without fine-tuning.
PythonPyTorchvLLMLLMsNLPAlignment
JasminTechs
StartupCo-Founder
Sep 2024 – Dec 2025
Waterloo, ON
- Co-founded a 0→1 AI tutoring platform that received positive feedback from 70%+ of testers.
- Stored course history for 50 active users in Aurora and archived recordings and documents in S3.
- Queued session summaries with SQS and sent notifications through SNS.
- Built Redis-backed session memory, cutting context retrieval latency by 80% from 120 ms to 24 ms.
VapiAWSAurora/RDSS3SQSSNSNext.jsFlaskRedis
Datacurve
LeadershipAmbassador
Jan 2026 – Present
Remote
- Connected with top ML engineers and researchers worldwide to collaborate on cutting-edge AI/ML projects and research initiatives.
- Represent Datacurve's community of elite ML practitioners — bridging talented engineers with high-impact projects in the ML space.
ML CommunityNetworkingAI Research
WAT.ai
LeadershipTechnical Project Manager
May 2025 – Jan 2026
Waterloo, ON
- Led a team of 9 building sentiment analysis and sarcasm detection models for stock market insights.
NLPSentiment AnalysisPythonProject Management
The GoldenHack
LeadershipDirector of Technology
Jun 2024 – Apr 2025
Waterloo, ON
- Designed and developed Waterloo's largest Business + Tech hackathon website.
- Collaborated with the team to innovate and improve the event experience.
Next.jsTypeScriptDesign
Akhny Solutions
InternshipSoftware Engineer Intern
May 2024 – Aug 2024
Milton, ON
- Designed mobile UI in Figma and implemented React Native interfaces, reducing navigation time by 40% and improving order completion rates.
- Developed Node.js + MongoDB backend with OAuth 2.0 authentication and AES/bcrypt encryption for secure order tracking and payment validation.
- Wrote 20+ unit and integration tests for backend APIs, catching regressions before release.
React NativeNode.jsMongoDBOAuth 2.0Figma
JuniorKids AI
InternshipSoftware AI Developer
Sep 2024 – Dec 2024
Montreal, QC
- Automated data collection every 30 minutes with AWS Lambda, storing structured data in PostgreSQL.
- Trained an LSTM model incorporating sentiment to forecast next-day prices, reducing out-of-sample RMSE by 5%.
PythonPyTorchAWSLangGraphPostgreSQL
Sep 2024 – Dec 2024
Waterloo, ON
- Developed a Chrome extension used by 50+ users to extract browsing data and build knowledge graphs for learning analysis.
- Integrated LLM-based recommendations to surface related concepts, improving learning speed by 35%.
Knowledge GraphHCINode.jsChrome ExtensionLLMs
Laurier Fintech
LeadershipAlgorithm Developer
Oct 2023 – Jun 2024
Waterloo, ON
- Spearheaded development of an open-source FinTech library, enhancing trading algorithms with in-depth backtesting for realistic market simulation.
PythonBacktestingFinTechAlgorithms
Uforse Education Group
TeachingComputer Science Instructor
Sep 2023 – Jan 2025
Ontario, Canada
- Designed and delivered online CS courses for high school students preparing for CCC and Euclid Contest.
- Covered data types, data structures, and algorithms; provided personalized mentoring and adapted curriculum to student needs.
PythonAlgorithmsData StructuresTeaching
Amazon
Almost ✓(｡•ˇ‸ˇ•｡)Software Development Engineer
Waitlisted
Canada
- Passed two interview rounds. Then gracefully ended up on the waitlist.
Projects
Things I've built.
Project
Jul 2025arXiv Paper Assistant
Chrome extension that enhances arXiv paper reading — AI-powered summaries, smart collections, and sidebar tools for researchers.
Chrome ExtensionWXTAI
View on GitHub →Project
Sep – Dec 2024UWPilot
Course planning tool for UWaterloo students — prerequisite tree visualization, QUEST schedule import, and academic pathway planning.
Next.jsKotlinStudent Tool
View on GitHub →Project
Nov 2024AI HealthBridge
Breaks language barriers for patients in Canada — real-time voice transcription, BERT NER on PubMED, and RAG with Llama-3.2-90B-Vision.
Meta HackathonLLMHealthcare
View on GitHub →Project
May – Oct 2024TrendSage
Correlates Reddit & 4Chan /biz sentiment with real stock trends using Apewisdom API, NLP models, and market data.
NLPPythonMongoDB
View on GitHub →Project
Aug 2024Fin4All
Democratizes investing — real-time stock analysis, Reddit sentiment insights, and personalized investment recommendations for all experience levels.
Hack the 6ixGCPLLM
View on GitHub →Project
Aug 2024MedScan Mentor
Helps medical students develop brain disease detection skills from CT scans with YOLOv5-powered real-time feedback.
YOLOv5Deep LearningHealthcare
View on GitHub →Research
NLP Applications, Safety & Alignment in LLMs.
Making large language models more ethically aligned and inclusive.
EMNLP 2025First Author2025
Pluralistic Alignment for Healthcare: A Role-Driven Framework
Jiayou Zhong et al.
A lightweight, role-based framework that generates structured ethical personas to simulate varied viewpoints. Evaluated on the VITAL benchmark, the method outperforms prior approaches across Overton, Steerable, and Distributional alignment settings — enabling scalable, fine-tuning-free alignment for inclusive and ethically sensitive AI.
NLPAlignmentHealthcareLLMsEMNLP 2025
Read paper →EMNLP 2026First Author2026
VISPA: Pluralistic Alignment via Automatic Value Selection and Activation
Jiayou Zhong et al.
Explores latent value alignment by steering hidden layer activations with controlled value vectors, enabling interpretable moral reasoning in large language models without fine-tuning.
NLPActivation SteeringValue AlignmentLLMsEMNLP 2026
Read paper →Blog.
Notes on finance, books, career moves, and life beyond work.
GitHub
Contributions in the last year.
A year of commits — research prototypes, side projects, and late-night debugging sessions.
695 contributions in 2025
LessMore
Off The Clock
K-dramas, soccer, and basketball.
The fun side. Because a site without personality is just a CV.
K-drama
What I watch
My Liberation Notes
All-time favorite
Can Love Be Translated
Currently watching
Soccer
Barca always
FC Barcelona through and through. The beautiful game at its finest — tiki-taka, pressing, and pure soccer.
Messi
The GOAT.
Pedri
My favorite current Barcelona player
Basketball
Steph Curry.
Warriors at home.
Best shooter ever. The Golden State Warriors are still the home team.
GOAT4× NBA Champion1× Finals MVP2× NBA MVP12× All-StarOlympic Gold