Huiru
Feng

About

Second-year Mathematics undergraduate at UCL.

At nineteen, I am studying mathematics in the UK as a Chinese student. My childhood in Shandong made me aware of the quiet script written for many women: study hard, find security, marry well, and shape a life around the expectations of others. While other children dreamed of meeting Prince Charming, my ambition was to create something of my own.

Data science, artificial intelligence, and the wider world of ideas hold a particular fascination for me. Answers alone rarely satisfy me. What matters is the reasoning beneath them and the freedom to question them.

Ambition becomes manageable through systems. I keep a taxonomy of my own mathematical errors, from skipped steps to final-line miscalculations to proofs argued in the wrong direction, and work through them one category at a time. I have come to trust a plan that can be revised more than one that has to be defended.

Writing, reflection, and tears are how I make sense of an inner world. Wisdom, to me, begins with the willingness to ask why a thought or feeling exists.

The woman I hope to become is brave, free, and grounded. She feels fear and doubt, and makes her choices anyway. She defines her life on her own terms. She can lose her footing, understand why, and stand again.

I am no bird; and no net ensnares me: I am a free human being with an independent will.
charlotte brontë · jane eyre

Academic

BSc Mathematics (G100), UCL · 2025–2028 · Year 1: First class, 75+

Year 1

• Real Analysis
• Abstract Algebra
• Number Theory
• Linear Algebra
• ODEs and PDEs
• Newtonian Mechanics
• Vector Calculus
• Probability
• Python

Year 2

• Complex Analysis
• Further Linear Algebra
• Metric Spaces
• Fourier Series
• Probability and Statistics
• Machine Reasoning
• Computational Methods
• Fluid Mechanics

Before UCL

2022 – 2025 · Jinan Foreign Language School, International Centre

A*A*A*A in Mathematics, Further Mathematics, Economics and Physics.

Where this is going

I plan to continue to a master's and then a PhD. For now that mostly means learning to do research properly rather than quickly. Reading the methods literature carefully, and getting used to writing down what I do not yet know.

Research

Both projects are in progress. Drafts available on request.

Robust factor decomposition for hierarchical consumer segmentation

Jul 2026 – present · with a PhD mentor

Standard k-means fails on high-dimensional consumer data for reasons that look like three separate problems but trace back to two design choices. Euclidean distance assumes dimensions are orthogonal and comparably scaled, so correlated features get counted more than once; and a single flat partition cannot express nested structure like "spending tier → price sensitivity".

I'm working on a pipeline that replaces the hand-set decisions, meaning thresholds, cluster count and factor weights, with parameters learned from the data: a robust loss in place of maximum likelihood, learnable factor weights under an L1 penalty, and two-level clustering so one result can be read at whichever granularity a decision needs. I wrote the abstract, introduction and methodology chapters.

Methodology drafted. Experiments not yet run, so there are no results to report.

Fallacy detection in UK parliamentary debate

Apr 2026 – present · with a PhD researcher at UCL · student journal, late 2026

A small team project on argument mining. The work started as multimodal fallacy detection and narrowed over three months to a text-based dataset.

My contributions: proposing the UK parliamentary questioning setting, identifying a workable label set, and building the annotated dataset using argumentation framework structures and IAT-style argument graphs. The main text is written by the PhD researcher leading the project.

Projects

Things I built, including one I'm taking apart again.

Algorithms and Big-O

May – Jun 2026 · Python, Matplotlib · UCL Year 1 summer project

I implemented bubble sort, merge sort, quicksort and counting sort from scratch in Python, without library shortcuts, so that the cost of every operation stayed visible in the code. I then designed timing experiments across input sizes from 100 to 10,000 elements and plotted the results on a logarithmic scale, which made the separation between O(n²), O(n log n) and O(n + k) growth something I could see rather than something I had been told.

The theory turned out to be the part I got most out of. I worked through the decision tree argument for the Ω(n log n) lower bound on comparison based sorting. Any algorithm restricted to pairwise comparisons has to distinguish between n! possible orderings, so its decision tree needs at least n! leaves, and a binary tree with n! leaves has depth at least log₂(n!), which is Θ(n log n). Counting sort escapes the bound because it never compares two elements with each other at all. It reads their values directly and uses them as indices, which only works when the range of values is known and bounded.

Writing up the proof in English rather than in symbols was harder than the implementation, and it was the first time I understood the difference between following an argument and being able to reconstruct it.

Cross-sectional alpha strategy, auto sector

Jan – Mar 2026 · team project, UCL Data Science Society · being rewritten

A cross-sectional ranking pipeline over nine global automotive equities. Raw prices were converted to daily returns to line up trading hours across time zones, momentum, volatility and RSI factors were ranked within each cross-section so that absolute levels became percentiles, and the model was a depth constrained Random Forest scored with Spearman rank IC rather than RMSE, because what matters is the ordering of the names and not the size of the error.

github · auto_stock_strategy

I am rewriting it from scratch. The reported Sharpe ratio was high enough that it almost certainly hides a bug, most likely look ahead bias or an absence of transaction costs, and I want to be able to say what every line does before I put my name on the result.

Strategic M&A valuation

Feb – Mar 2026 · cross-university team (UCL, Cambridge, LSE, Imperial)

Our team acted for Samsung Electronics and had to pick an acquisition target in the semiconductor sector, then defend the price we put on it. We screened candidates and settled on Lattice Semiconductor, a designer of low power FPGAs, on the grounds that Samsung had almost no presence in programmable logic while its foundry share sat near 9% against TSMC's 64%.

I led the quantitative work. I built a five year forecast from management guidance, with revenue growth tapering from 37% to 15% and EBIT margin recovering from 15% toward 28% as operating leverage returned. WACC came out at 11.45% using CAPM with a beta of 1.67 and a risk free rate of 4.04%, alongside a 2.5% terminal growth rate and an 18% normalised tax rate.

What I remember most is that the three methods disagreed with each other. DCF gave a floor of $22 to $35, EV/Revenue gave $28 to $35, and forward EV/EBITDA gave $60 to $137. We recommended $127 per share, a 30% premium to the $98 market price. Most of that spread comes from things a spreadsheet cannot produce: control of the asset, the HARMAN automotive integration, and the case for bundling Samsung HBM memory with Lattice programmable logic in the same AI server. Learning to write the argument for why a premium is defensible taught me more than building the model did.

AI wearable concept

Apr 2026 · 7-person cross-university team

A seven person team from UCL, Oxford and Bristol working on a companion wearable for anxiety detection in university students. I handled the market side, which meant sizing the segment, mapping the existing products and where they fell short, and working out how a device like this would actually reach the people it was designed for.

It was the first time I had to argue for something without a proof behind it, and I found the ambiguity harder than any of the maths.

Career

Open to summer 2027 internships.

Algorithm Team Intern · Inspur Group

Aug 2026 – present · Jinan, China

Working under an algorithm mentor: extracting and restructuring SQL query output for analysis, migrating a MySQL/SQLite schema to GaussDB, and sourcing container images for a YOLO deployment. Early days, and mostly a matter of learning how the infrastructure fits together.

Front-end Development Intern · Inspur Group

Jun – Aug 2026 · Jinan, China

Worked under a mentor on a WeChat mini-program for staff canteen ordering, covering cart, order lifecycle, merchant fulfilment and menu management. The implementation leaned heavily on AI-assisted coding; my contribution was finding where the generated logic broke, in pricing precision and in order state transitions, and driving it to a working build. First hands-on experience with git, front-end/back-end separation and JSON.

Head of Project Labs · UCL Data Science Society

Mar 2026 – present · London

Elected to lead the society's 2026/27 project programme, setting academic direction across data science, machine learning and quantitative analytics, and running the project teams from scoping through to final presentation. The programme starts this term.

Maths Teaching Assistant · Lumi Summer School

Jul 2025

Marked assignments and exam papers, taught TMUA and MAT past papers, and ran Oxbridge-style mock interviews.

Finance and Strategy Mentor · FRC Team 6399

Aug 2025 – Mar 2026 · remote

Managed the team's competition budget and expense records across two competitions, in Shanghai and in Turkey.

Journal

Fragments I decided not to leave buried in Notes. Open only the one that catches.

The rest of me

Music, books, screens, and a lot of airports.

On repeat

I saw TWICE live in London in 2026, five years after I started listening.

Screen

• All 34
• Korean drama
• Film
• Series
• Chinese drama

Reading

Not a long list. I read slowly and reread the ones that unsettled me.

Places

countries visitedChina, by province