I'm a student developer working at the intersection of applied mathematics, probabilistic modelling, and scientific open-source engineering. I build systems that make uncertainty, failure conditions, and provenance explicit — then test whether conclusions hold outside a clean demo.
The common thread across everything I build: replace impressive-looking outputs with calibrated evidence.
Quantitative Modeling & Data Analysis intern at Energy Innovation: Policy & Technology LLC (Summer 2025) — electricity demand forecasting, extreme-weather calibration, tail-risk analysis under Lele Zou.
| Competition | Result | What it means |
|---|---|---|
| AIME 2026 (American Invitational Mathematics Examination) | 14 / 15 | Invitation-only second round of the US Math Olympiad pipeline (AMC 10/12 → AIME → USAMO/USAJMO → IMO team). Only a small fraction of AMC 10/12 takers qualify (MAA guarantees at least the top 2.5% of AMC 10 and top 5% of AMC 12). A 14 is approx. top ~2.5% of AIME takers — roughly top ~0.25% of all AMC participants (approximate, my own estimate). |
| USAJMO 2026 (USA Junior Mathematical Olympiad) | Qualifier | Invitation-only USA Junior Math Olympiad — the next stage after AIME. |
| British Mathematical Olympiad Round 2 (BMO2) 2026 | 27 / 40 — Distinction · Best in School | BMO2 is the final olympiad round used to select the UK International Mathematical Olympiad (IMO) squad. As an international student I was not officially ranked; based on UKMT's published BMO2 2026 score distribution (8 students scored 26+), my score corresponds to about 8th overall (international, unranked). |
| Stanford Math Tournament (SMT) 2026 | Algebra — 3rd place · Discrete — 7th place | Subject-test placements at one of the largest university-run math contests (~800 in-person competitors; 2,000+ participants globally including SMT Online, per SMT). |
|
EPSILON — Algorithmic Trading Decision Lab Falsifiability-first quantitative research environment. Failure conditions are locked in before results appear. Realistic order types, slippage, FFT analysis, robustness diagnostics. Audited 200+ LightGBM fits — detected leakage, IC dropped 0.94 → 0.10. Strategy rejected. Diamond Challenge Beijing Regional — 3rd place.
|
Low-cost AI glasses for visually impaired users. Multimodal: traffic signals, obstacle detection, voice feedback, computer vision. Safety-first design — low-confidence detections never trigger action commands. Deployed and donated across three communities; taught older adults to self-assemble devices.
|
23 external pull requests merged across 14 organizations.
| 23 merged PRs |
14 organizations |
15 repositories |
Contributions to: HuggingFace/peft · langgenius/dify · networkx/networkx · fairlearn/fairlearn · Nixtla/statsforecast · Nixtla/neuralforecast · SALib/SALib · beeware/briefcase · scikit-lego · scikit-learn-contrib/MAPIE · argoproj/argo-cd · ag2ai/faststream · and more
Four representative contributions
| PR | What it did |
|---|---|
| MAPIE #953 | Added AUROC / AUARC uncertainty metrics; 349 additions across 5 files |
| Fairlearn #1674 | Confidence-interval support in plot_metric_frame via public MetricFrame API |
| NetworkX #8735 | Regression-tested warning for ineffective PageRank convergence threshold |
| BeeWare Briefcase #2939 | Hardened 27 GitHub Actions refs; pinned permissions; zizmor scanning |
→ Full evidence record with dated methodology
PJM Polar-Vortex Load Forecasting (in progress) — Annual 90% interval coverage 82–85%, collapsing to 31–33% during the 2014 cold event. 78,888 point forecasts. Demonstrates how standard annual validation conceals systematic failure during rare extremes.
The Backtest That Lied — 200+ LightGBM fits audited. Leakage detection cut information coefficient from 0.94 → 0.10. Strategy rejected after costs and liquidity constraints.
→ Research notes · → PJM paper repository
| Competition | Result |
|---|---|
| AIME 2026 | 14 / 15 — Qualification |
| British Mathematical Olympiad Round 2 | 27 / 40 — Distinction · Best in School |
| Stanford Math Tournament 2026 — Algebra | 3rd place |
| Stanford Math Tournament 2026 — Discrete | 7th place |
| British Physics Olympiad Round 1 2025–26 | 80 / 100 — Top Gold · Beijing #5 |
| UK Senior Mathematical Challenge | 120 / 125 |
| LSESU Economics Challenge | 97 / 100 |
| World Scholar's Cup Beijing 2025 | Overall Champion Scholar; Senior Team 3rd |
Contribution counts verified via GitHub API, November 2026. Calibration and methodology notes: evidence record.