SDSC 2005 · Introduction to Computational Social Science · City University of Hong Kong

An early warning system for US visa-issuance surges that fuses 170k+ news articles, Google Trends interest and real exchange-rate signals across 15 origin countries into one panel. News is embedded, clustered and labelled on quantized TensorRT engines, and a horizon-aware ensemble forecasts surges one to six months ahead.

• Walk-forward F1 0.96 at a 1-month lead, 0.86 at 6 months; recall never below 0.92
• 170,784 articles crawled and filtered to 104,333, clustered with cuML HDBSCAN
• Jina v5 embeddings at ~20k tokens/s via TensorRT INT4 (cosine 0.9999 vs FP32)
• 58 of 392 country-event-lag signals survive Benjamini-Hochberg correction

[Report (PDF)] [Code] [Datasets]

Streamlit and Plotly dashboard with LLM-backed signals

Fuses ten years of NVDA price history with 7,000+ collected news articles into sentiment-driven signals, technical overlays and hypothetical trade analysis. Sentiment runs through FinBERT and VADER, the technical layer through the indicators-cli package, and the narrative layer through an LLM.

• Market data, news sentiment, technical indicators and trade simulation in one view

[Code]

Open-source command-line tool, published on PyPI

Fetches historical prices from Yahoo Finance and computes nine technical indicators across four timeframes, for one ticker or a batch, writing CSV, JSON, Parquet or Excel. Packaged, documented and released on PyPI, and reused as a dependency by the stock dashboard.

• 9 indicators (SMA, EMA, MACD, RSI, Bollinger, ATR, OBV, ROC, Stochastic)
• Batch tickers, custom configs, templates and optional GPU path

[PyPI] [Code]

Open-source Kubernetes and Kafka deployment

Live IoT sensor data is scarce and usually paywalled. This project replaces it with a configurable Rust producer/consumer fleet on Kafka, deployed to Kubernetes through ConfigMaps, so a simulation can be scaled to any stream rate and wired into other microservices.

[Code]

Tracked, reproducible classification experiments

An end-to-end pipeline over 5,574 labelled SMS messages that compares five classifiers, with every run's hyperparameters, metrics and artefacts tracked in MLflow. The best model separates scams at 94% accuracy with a low false-positive rate.

• 5,574 messages, 5 model families, MLflow experiment tracking

[Code]

GE2260 · Introduction to Finance · City University of Hong Kong

Constructs a 60/40 fixed-income and growth portfolio from five years of price history, then locates the efficient frontier with Markowitz mean-variance optimization and 20,000 Monte Carlo paths. The growth sleeve prices 18.3% expected annual return at 11.5% volatility, and the blended portfolio is projected to beat the S&P 500 on return and drawdown alike.

• 20,000 Monte Carlo simulations; collar hedge for high-volatility regimes
• Diversification verified across a five-year correlation matrix

[Report (PDF)] [Video] [Code]