Liang (Leon) Wu
AI Engineer · Quantitative Developer
10+ years building deep-learning systems, from research models to production mobile inference, now applied to quantitative finance: agentic LLM trading frameworks, ML trading strategies, market-data pipelines and backtesting infrastructure in Python & C++.
Hong Kong
MSc Financial Engineering · MPhil ECE · MPhil CS
10+ years in AI
Projects AI × Quantitative Finance
Freqtrade Agent
Researcher→Quant Developer→Risk Reviewer→Human Approval→Paper / Live Bot↺Monitor
- Agent research pipeline on a DeepSeek harness: a Researcher explores market data and proposes testable ideas, a Quant Developer implements them as Freqtrade strategies and iterates on backtests, and an independent Risk Reviewer checks for overfitting and logic flaws.
- Scheduled automation: daily research runs (e.g. BTC/USDT, ETH/USDT trend and mean-reversion search under drawdown constraints) and nightly Monitor runs that inspect every paper/live bot and flag anomalies. Each agent has its own backtest and validation budget; every run is logged with token usage.
- ML strategies: logistic-regression and gradient-boosted signals, FreqAI for PyTorch / Transformer models, multi-horizon return/volatility/volume features, rolling retraining, label-availability checks and holdout validation.
- Trading platform: FastAPI + Vue console covering strategies, backtests, paper/live trading on Binance, a live-trading approval queue, accounts, market data and a task queue, with risk controls and audit trails throughout.
A-Share Live Trading System
- Daily automated pipeline: a scheduler pulls TuShare data before the open (07:00), runs stock screening and position sizing, and writes the day's trading plan.
- Configurable screening & sizing: number of holdings, PE and market-cap floors, sort order; total-capital, cap-ratio and cash-ratio sizing, with manual overrides.
- Performance analytics GUI: interactive equity & return curve against SSE Composite, CSI 300 and ChiNext, with max-drawdown highlighting and daily P/L inspection.
Stock Backtesting Platform
backtest · 2020-01 → 2026-03 · 1,489 trading days × 4,581 stocks · net of fees · CSI 300 benchmark
- Data layer: TuShare, AKShare and yfinance normalized into MySQL, with a cached full-market panel (~1.3 GB) that loads in seconds for repeated experiments.
- Backtest engine: pluggable strategies with configurable capital, max holdings, ranking order and sizing; realistic A-share costs (minimum commission, commission rate, stamp duty, transfer fee, slippage).
- Analytics: return vs CSI 300, portfolio value, drawdown curve, daily-return distribution, plus Sharpe, volatility, average position and trade statistics.
TradingAgents-FuTu
- Extended LangGraph-based TradingAgents with Futu/moomoo data for HK/US prices, indicators, fundamentals and news.
- Vendor routing with Yahoo Finance fallback, symbol normalization, OpenD error handling; date-filtered reports to prevent look-ahead leakage.
- 104 automated tests with mocked SDK/HTTP backends, so CI runs without a live OpenD gateway.
DeepSeek Agent Harness
- Multi-step tool-use workflows across browser automation, file system and external APIs.
- State tracking, error handling and human-in-the-loop checkpoints; market-data and research tools wired into a stateful multi-agent trading workflow.
Skills
- AI & Deep Learning
- PyTorch, Transformers, CNNs, BERT, CLIP, vision-language models; model optimization and real-time deployment with ONNX, NCNN, MNN, TNN.
- Agentic Systems
- LangChain, LangGraph, DeepSeek agent harness, multi-agent orchestration, stateful workflows, tool integration, structured outputs, human-in-the-loop.
- Quantitative Finance
- Options pricing, financial econometrics, portfolio backtesting, algorithmic strategy design & evaluation; ML signals, rolling retraining, holdout validation.
- Trading Data & Engineering
- Python, C++, SQL/MySQL, Linux, Git; FastAPI, Vue, Freqtrade, FreqAI, Futu OpenAPI, AKShare, TuShare, yfinance, Backtrader, vn.py.
Experience
Mobile Deep Learning Model Developer · Tap Mobile (Contract)
2020 – 2026 · Remote- Designed, trained, optimized and deployed deep-learning models in mobile apps serving 1M+ users.
- Transformer-based face-parsing and image-inpainting models optimized for real-time mobile inference.
- Fast document-dewarping model in production; hybrid CNN-Transformer four-point document detector with 97%+ mIoU in real time.
Researcher · Hong Kong University of Science and Technology
2019 – 2024 · Hong Kong- Transformer-based video expression recognition: 4th place in the ABAW challenge (CVPR Workshops 2023).
- Video-based gaze estimation with state-of-the-art performance.
- Multimodal gaze interface for controlling wheelchairs and teleoperated robots.
Perception Team Lead · Allride Inc.
2018 – 2019- Led point-cloud perception for vehicle & pedestrian detection, including PointPillars adaptation; technical direction for the team.
Image Algorithm Researcher · ReadSense Inc.
2016 – 2018- YOLO- and MobileNet-based object detectors optimized for real-time inference on mobile devices.
Image Algorithm Engineer · Alibaba Inc.
2015 – 2016- LSTM-based OCR for ID-card text recognition deployed in Alipay; OCR and image classification for Taobao product images.
Research
I received the M.Phil. degree from the Department of Electronic and Computer Engineering at the Hong Kong University of Science and Technology, under the supervision of Prof. Bert Shi. My research interests focus on computer vision, machine learning and robotics. Previously, I received another M.Phil. in Computer Science from Nanjing University, China, supervised by Prof. Tong Lu, where my research mainly focused on text detection and recognition in images and videos.
Computer VisionMachine LearningRoboticsGaze EstimationAffective Computing
Publications * equal contribution
Education
M.Sc. in Financial Engineering · City University of Hong Kong
2024 – 2025- College of Business. Options pricing, corporate finance, financial econometrics, algorithmic trading.
M.Phil. in Electronic & Computer Engineering · HKUST
2019 – 2024- Computer vision & deep learning. Supervisor: Prof. Bertram Shi.
M.Phil. in Computer Science · Nanjing University
2012 – 2015- Image processing & machine learning. Supervisor: Prof. Tong Lu.
Awards
- 2015 Excellent Master's Degree Thesis, Nanjing University (3 of 140+)
- 2015 Outstanding Graduate, Nanjing University (17 of 140+)
- 2015 Excellent Postgraduate, Nanjing University
- 2014 Excellent Postgraduate Scholarship, Nanjing University