Projects

💻 AI & Algo Trading Strategies

Role: AI Algorithm Intern @ Baidu Qianfan

Challenge: Traditional time-series models (Transformers) struggle to capture high-order correlations between stocks in a non-stationary market.
Solution: Developed Meta-HCapsNet (Hypergraph Capsule Networks). Designed a sparse dynamic routing mechanism to solve the cold-start problem in small-sample scenarios.
Result: Outperformed SOTA baselines (iTransformer) by reducing MAE by 54% in real-market backtesting.
Tech Stack: PyTorch, Hypergraph Learning, Capsule Networks.

Role: Team Leader & Strategy Developer

Overview: Designed an institutional-grade ETF portfolio strategy for the “SSE Cup” (organized by Shanghai Stock Exchange), winning the Grand Prize (National 1st) out of thousands of teams.
Strategy: Implemented a “Core-Satellite” framework driven by structural market trends:
- Conservative Portfolio: Constructed a “Low-Vol Core + High-Beta Satellite” mix. Achieved 10.55% monthly return with a Sharpe Ratio of 16.90.
- Aggressive Portfolio: Focused on high-growth sectors (Semiconductors & Aerospace). Achieved 18.71% monthly return with a Max Drawdown of only 1.1%.
Tech Stack: Python (Backtrader), Mean-Variance Optimization, Risk Management.

Role: Product Manager & Lead Developer

Overview: A gamified asset allocation mini-program embedded in the Ping An Securities App, designed to solve Gen-Z’s investment anxiety.
Innovation: Gamified user profiling (Tank/Marksman/Mage) to match risk preferences. Visualized “Peer Consensus” using a Trend Radar to reduce decision paralysis.
Algo Engine: Built a backend engine integrating Brinson Attribution for performance analysis and Monte Carlo Simulation (10k+ runs) for stress testing.
Achievement: Won 1st Prize in the Ping An Securities Financial Literacy Competition.
Tech Stack: React 18, Python (FastAPI), PostgreSQL, Risk Parity Model.

Role: Research Assistant

Task: Built an automated ETL (Extract, Transform, Load) pipeline to process large-scale unstructured Chinese fiscal reports.
Implementation: Used OCR to parse thousands of PDFs and developed a rule engine to extract 16-dimensional fiscal factors.
Outcome: Constructed a comprehensive micro-database for intergovernmental transfer payments to support national-level economic research.
Tech Stack: Python (Pandas), OCR, SQL, Regular Expressions.

Role: Founder & Producer

Impact: Achieved 48M+ streams and 23k+ followers on NetEase Cloud Music (Top 5% of creators).
Work: Managed a team of creators, produced 20+ originals and 450+ covers.
Link: Listen on NetEase