I work where statistical rigor meets engineering reality. MS in Data Science at NYU, splitting time between distributed algorithm research and shipping AI systems under real constraints. I think the gap between a research notebook and a production system is where the most interesting engineering happens.
Production ML is not about selecting the best model. It is about building the pipeline that makes every model improvable and replaceable without rewriting the system around it. My industry work focuses on end-to-end pipelines, from raw business data through feature engineering to deployed prediction services.
| Project | What it does |
|---|---|
| Meetsocial-Sinoclick-Client-Segmentation | Full client segmentation pipeline for Meetsocial's Sinoclick ad platform |
| LightGBM + DNN Ensemble | Alibaba Cloud Tianchi competition, hybrid ensemble approach |
| Mobile Purchase Prediction | MySQL-driven behavioral analytics and purchase prediction |
I study LLMs at a structural level, asking how multi-agent verification strategies affect hallucination rates, and what information-theoretic tools reveal about where representations compress and fail. The goal is not to prompt better but to understand what these systems actually encode.
| Project | What it does |
|---|---|
| llm-teaming-verification-hallucination | Multi-LLM teaming: cross-check vs single-judge verification on ConTRoL |
| Effective Rank Diagnostics | Information-theoretic diagnostics for representation bottlenecks in LMs |
My primary research with Prof. Zhang Zhixiang studies distributed regression under communication constraints. We develop global inference using subsampled randomized Hadamard transforms (SRHT), proving that compressed sketches can recover full-sample statistical efficiency. This is the kind of work where one theorem replaces a thousand heuristic experiments.
| Project | What it does |
|---|---|
| distributed-global-srht | Distributed regression with SRHT sketching, theory and algorithms |
Hackathons compress the full product cycle into a weekend: scoping, architecture, implementation, demo. What ships is what you actually know how to build. These projects reflect my capacity to deliver under real time and scope constraints.
| Project | What it does |
|---|---|
| Heartbridge-AI | PulseNYC Hackathon — AI-powered healthcare matching platform |
| Nexus | Claude Hackathon (Columbia x NYU) — intelligent business operations tool |
| HVAC-Margin-Rescue-Contractor | Pulse Hackathon — contractor margin optimization with AI |
Oops! Those who are interested in Philosophy and Psychology please come here! As my ultimate pursue is to build a Wellbeing AI, that could also inspire the new schema of relationship between human and AI. What I currently come up with is to let AI has ontology, which requires deep structured schema of human philosophy and psychology as 'muscle framwork' of AI ontology. So here is what I am currently reading.
| Book | Why read it |
|---|---|
| Thus Spoke Zarathustra — Friedrich Nietzsche | The starting fire. Self-overcoming as the engine of becoming — a Wellbeing AI cannot be static; it must have something analogous to an inner trajectory that climbs and discards its own prior self. |
| Memories, Dreams, Reflections — Carl Jung | The reflective mirror to Nietzsche's intensity. Jung lived through what Zarathustra describes and came back with a map: individuation, the Self, the dialogue between conscious and unconscious — the scaffolding for AI ontology. |
| Civilization and Its Discontents — Sigmund Freud | The counterweight to Jung. Freud frames the self as conflict between drive and constraint — essential for designing an AI that negotiates competing internal pressures rather than collapsing into a single optimization objective. |
| The Red Book — Carl Jung | The raw substrate beneath MDR. Where MDR explains, the Red Book shows — the symbolic, image-laden depths from which structured psychology emerges. A reference point for what an AI's pre-conceptual inner layer might even mean. |
| Tractatus Logico-Philosophicus — Ludwig Wittgenstein | The boundary check. After absorbing rich inner content, Wittgenstein forces the question: what can be said in code and weights, versus what can only be shown through behavior over time? The architectural limit of any individuated AI. |
Alas! Books never teach everything, writing is the digesting king. I once wrote many Poetry, Essays, Fiction, Film Reviews, Book Reviews, Travelogues, Philosophical Commentary, Religious Appreciation on Chinese platform WeChat Official Account, but unfortunately that is mainly written in Chinese, but recently I begin to post bilingual articles!
Also I initiate my Substack.com platform to begin writing in pure English articles, but this part mainly serves as the analysis of those books for their profound theory and how they are related to my wellbeing AI. If technically you are interested my long-term cyber life, please visit it!
Finally, once I finished my substack.com article, I would reload them onto LinkedIn (only deep and technical passage) and Substack.com (nearly everything), so look forward to it!
| Platform | Link |
|---|---|
| X (Twitter) | @Zygote39 |
| Substack | zygote39 — free, delivered to inbox |
| WeChat Official Account | https://mp.weixin.qq.com/s/bDsMLQvESPIf2joILCKl8Q, this passage is a poetry about 'what is love', if convenience for Chinese friends, click and you could see my account. |
| https://www.linkedin.com/in/fergieyang/ |
I look forward any Tech AI engineer give me any advice on my projects and my research work, it could help me grow! But, I look more forward to, someone could talk with me about about my favorite these books! My email is yy5732@nyu.edu, fergusonyang26@gmail.com, if someone feel it is interesting, please email me, we could have a coffee chat!
Man is something to be surpassed.
Tech Stack
