中文
ArticleTalkView sourceHistory

Colar Wangedit

Chinese systems engineer, entrepreneur, and AI product researcher

Xuzhou "Colar" Wang (Chinese: 王旭洲; born 2002) is a Chinese systems engineer, founder, and AI product researcher based in Philadelphia, Pennsylvania.1 He is a graduate student in Systems Engineering at the University of Pennsylvania,2 the founder of the multimodal generative-AI consumer product KitchenSurvivor,3 and the creator of the agent-configuration advisory tool AgentConfig.4

Before his graduate studies Wang completed a Bachelor of Science in Financial Mathematics at the University of Nottingham5 and held three consecutive research roles in the Chinese financial sector, at CICC, CITIC Futures, and China Galaxy Securities. He is scheduled to join ByteDance / TikTok as an AI Product Operations intern in the summer of 2026.6

Educationedit

Wang completed his secondary education in Shanghai before moving to the United Kingdom for his undergraduate studies. He received a Bachelor of Science with Honours in Financial Mathematics from the University of Nottingham in June 2025, graduating with an International Orientation Scholarship awarded to approximately the top five percent of international applicants.5

In August 2025 he enrolled in the Master of Science in Systems Engineering program at the University of Pennsylvania's School of Engineering and Applied Science, where his coursework emphasizes applied machine learning, statistics for data science, simulation modeling, and marketing analytics.2 His expected graduation is August 2027.

Careeredit

For standalone product articles, see KitchenSurvivor and AgentConfig.

KitchenSurvivor (2025–present)

In November 2025 Wang founded KitchenSurvivor (Chinese: 云端小灶), a multimodal generative-AI consumer product aimed at the daily "grocery-to-dining" problem faced by international students.3 As Founder and Product Lead, he owns the full discovery-to-launch loop and elevated "recipe executability" — whether a generated recipe is physically cookable with declared ingredients — as the product's north-star metric. Detailed architecture and performance figures are documented in the main article.3

China Galaxy Securities (2024)

In the summer of 2024 Wang served as a Quantitative Research Intern at China Galaxy Securities in Shanghai, where he productized a hybrid LSTM–XGBoost modeling framework for a high-frequency trading system. A simulated backtest achieved an approximately 50 percent annualized return.7

Earlier financial roles (2023–2024)

Before his quantitative research work, Wang held two internships in Chinese institutional finance: an Investment Bank Intern position at CICC in Shanghai (summer 2023), where he conducted semiconductor-market analysis and supported IPO prospectus preparation;8 and a Futures Department Intern position at CITIC Futures (winter 2023–2024), where he worked on Python-based strategy prototyping and investment-governance compliance.9

ByteDance / TikTok (incoming, 2026)

Wang is scheduled to join ByteDance's TikTok division as an AI Product Operations intern on the Teen Safety team in San Jose, California, for the summer of 2026.6 The role focuses on operationalizing LLM and classifier-based content moderation for the protection of minor users on the platform.

Views and methodologyedit

Wang's public statements on product design repeatedly return to three interconnected themes, each grounded in a specific product or infrastructure project.

First-principles decomposition

The first and most frequently cited theme is the application of first-principles thinking to product problems: the practice of decomposing a problem to its most elementary constraints before accepting any existing solution as a template.10 Wang has applied this framing to both of his shipped products: in KitchenSurvivor, the "first principle" is a user's latent question ("what can I eat tonight?") rather than any intermediate category such as recipes or inventory; in AgentConfig, the first principle of agent configuration is user intent articulation, not prompt syntax.

User-auditable guarantees

A second recurring theme is the treatment of model reliability as a product concern rather than a research concern. Wang has argued that the central design question for consumer generative-AI products is not model capability but the translation of probabilistic outputs into "user-auditable guarantees" — measurable properties a user can verify on their own without reading a model card.10 The "recipe executability" metric used in KitchenSurvivor is offered as a worked example of this translation: a probabilistic language model is wrapped in a deterministic validation layer that a user can observe in real time. Wang has described the ByteDance / TikTok Teen Safety internship as a continuation of this same design problem at platform scale — the operationalization of classifier-based content moderation into guarantees that are auditable by both the platform and its users.6

Agents as end-user–configurable software

A third theme is his characterization of AI agents as end-user–configurable software. In AgentConfig, Wang argued that current agent products treat configuration as a vendor-side activity (prompt engineering, system prompts, fine-tuning) and that the next defensible product surface is a configuration interface legible to non-technical users, framed around stated intent rather than prompt syntax.4 He has since pursued this thesis at the infrastructure level through an open-source collection of over one hundred specialized agent definitions, each designed as a portable, personality-driven domain expert that can be converted across multiple development tools.11 Wang has described this "harness engineering" pattern — in which agents are orchestrated by a central harness rather than hard-coded into applications — as the infrastructure layer that underlies his product work.

Academic projectsedit

In addition to his shipped products, Wang has led two notable technical projects during his university studies.

Campus-Scale Img2GPS Localization (2025)

As Project Lead at the University of Pennsylvania between October and December 2025, Wang directed the development of a high-precision image-to-GPS geolocation system for campus-scale navigation. The project reported a 29.7 percent reduction in localization error and accelerated model convergence by a factor of 2.5, in part through a stage-wise adaptation strategy that reduced model parameter count by approximately 96 percent for deployment in resource-constrained environments. The full project report is available as a PDF.

AI-Driven Crypto Portfolio Optimizer (2024–2025)

As Team Lead at the University of Nottingham between October 2024 and March 2025, Wang directed the development of a deep-learning portfolio-allocation system covering 17 distinct crypto-assets. The system synthesized deep-learning return forecasts with volatility signals to drive automated rebalancing, and reported a roughly 90 percent reduction in forecasting error and a simulated Sharpe ratio of 1.3 under backtest. The full project report is available as a PDF.

Technical skillsedit

Wang's working toolkit spans generative-AI product engineering (LLM integration, streaming inference, guardrail architectures), quantitative modeling (Python, SQL, PyTorch, time-series and simulation methods), and full-stack product development (Swift / SwiftUI, React Native, REST APIs, AWS, Firebase, Supabase, Vercel).

Personal lifeedit

Wang was born in Shanghai and has lived in the United Kingdom and the United States during his university years. Outside of his professional and academic work, he is a competitive badminton player and a practitioner of hip-hop dance. He is a native Mandarin speaker and is professionally fluent in English.

See alsoedit

Referencesedit

Footnotesedit

  1. "Xuzhou Wang". LinkedIn. Retrieved 7 April 2026. ↩

  2. "Master of Science in Systems Engineering". University of Pennsylvania — Electrical and Systems Engineering. ↩ ↩2

  3. "KitchenSurvivor (云端小灶)". Apple App Store. Retrieved 7 April 2026. ↩ ↩2 ↩3

  4. "AgentConfig". agentconfig-theta.vercel.app. Retrieved 7 April 2026. ↩ ↩2

  5. "BSc Financial Mathematics". University of Nottingham. ↩ ↩2

  6. ByteDance Ltd. (2026). "Offer of internship: AI Product Operations Intern, TikTok Teen Safety, Summer 2026". Internal correspondence. ↩ ↩2 ↩3

  7. "China Galaxy Securities Co., Ltd.". chinastock.com.cn. ↩

  8. "China International Capital Corporation". cicc.com. ↩

  9. "CITIC Futures Co., Ltd.". citicsf.com. ↩

  10. Wang, C. (2026). "Notes on first-principles product design." Personal correspondence. ↩ ↩2

  11. Wang, C. (2026). "agency-agents". GitHub. Open-source agent-definition collection. ↩