LI Junxing

李俊星 Jasen

العربية · EN · Español · فارسی · Français · हिन्दी · 日本語 · 한국어 · Русский · 中文

Welcome to this site! This page contains more details about me.

🙋 Profile

Research engineer with a mathematics and statistics background, working across quantitative research engineering, financial data infrastructure, computer-vision post-processing, and enterprise AI-agent delivery. I can take a problem from definition and experimental validation through reproducible data/model workflows, engineering delivery, and user-facing documentation.

My current practice combines quantitative data platforms and signal research with RankSEG image-segmentation research engineering and practical AI-agent workflow design. I care about systems that are testable, repeatable, and useful in real operating environments.

Current focus: quantitative research engineering, financial data infrastructure, research platforms, trading-data services, and AI solutions / agent automation.

🌈 Color Balance

Color   Share   Bar            Meaning
-----   -----   ------------   ---------------------------------------------
Black   40.6%   ████████░░░░   Agency, achievement, strategic execution
Blue    26.9%   █████░░░░░░░   Understanding, mastery, analytical depth
Red     14.4%   ███░░░░░░░░░   Intensity, candor, pressure tolerance
White   12.7%   ███░░░░░░░░░   Structure, order, stability
Green    5.4%   █░░░░░░░░░░░   Connection, emotional attunement, harmony

My profile is strongly Black-Blue: action paired with analysis.

• Black shows up as drive, agency, and a bias toward getting things done.
• Blue adds precision, curiosity, and a preference for understanding how systems actually work.
• Red contributes some appetite for pressure, directness, and intensity.
• White is present, but secondary: I value structure when it improves execution.
• Green is the least prominent, which means I am less naturally oriented around emotional atmosphere or consensus-building.

Core Pattern

I do my best work when there is a difficult problem, a real constraint, and enough room to think clearly.

I am drawn to environments where rigor matters, where complexity is not avoided, and where outcomes can be judged by whether they hold up in practice. I tend to trust methods that are testable, repeatable, and grounded in reality. I would rather build a working system than produce a persuasive description of one.

In teams, this often means I contribute through problem-solving, structure, and execution. I may not always lead with warmth or visibility, but I tend to care by making things work, reducing ambiguity, and carrying difficult tasks across the finish line.

Reference

• SoulTrace Result: https://soultrace.app/en/results/48093f7c-7980-4f83-97a2-e5b8df3a103f

🎓 Education

• City University of Hong Kong (Sep 2025 – Oct 2026)
  Master of Science in Statistics

• Qingdao University of Technology (Sep 2020 – Jun 2024)
  Bachelor of Science in Mathematics and Applied Mathematics

💼 Internship & Research Experience

Sika China

AI Agent Intern | Apr 2026 – Jul 2026

• Requirements decomposition and delivery: Worked across AI product support, internal solutions, data automation, and business operations for repetitive, experience-dependent workflows.
• Agent/skill design: Turned loosely defined business needs into structured inputs, rules, MVPs, feedback loops, and acceptance criteria.
• Internal rollout: Supported agent enablement, user training, SOP documentation, and reusable workflow assets while keeping business information within appropriate boundaries.
• Business automation: Built and adapted practical automations for data processing, knowledge retrieval, reporting, and presentation workflows, balancing visual quality, editability, and reuse.

The Chinese University of Hong Kong (CUHK)

Part-time Research Assistant | Feb 2026 – Present

• Research engineering: Drove research engineering and deployment of the RankSEG image-segmentation post-processing algorithm, covering reproduction, framework adaptation, inference integration, evaluation, and cluster packaging.
• Framework integration: Advanced RankSEG across mmsegmentation, PaddleSeg, Hugging Face Transformers, and SAM-family paths; designed interfaces for Dice/IoU-oriented post-processing without retraining.
• Reproducible evaluation: Built evaluation, logging, and result-tracking workflows to improve comparability, debugging, and reproduction stability.
• HPC packaging: Packaged Singularity + Slurm environments on CUHK HPC and prepared reusable multi-GPU job templates.
• Open-source collaboration: Contributed through merged Transformers PR #19, SAM integration PR #20, integration documentation PR #22, and MONAI PR #2.

Zhejiang Mingce Asset Management Co., Ltd.

Quantitative Architecture Intern | Sep 2025 – Apr 2026 Quantitative research engineering and digital-asset data infrastructure

• Data platform: Built a quantitative research platform covering about 8 years of history and the liquidity Top 40 digital assets, integrating OHLCV data, on-chain indicators, and derived factors.
• Columnar architecture: Managed roughly 170 million minute-level records, 3,000 candidate features, and about 20 GB of Parquet data; reworked storage around Parquet, S3-compatible object storage, DuckDB/Polars, and MySQL metadata.
• Query and refresh optimization: Added partitioning, incremental collection, validation, and backups, reducing typical research queries from 1–3 minutes to 5–15 seconds and keeping daily updates within 10 minutes.
• Research pipeline: Refactored scattered Python/R work into a configurable Python–R–SQL pipeline for multi-asset training, inference, rolling validation, backtesting, and reproducible experiments.
• Model validation: Used LightGBM, XGBoost, MAVE, and VAE for 7/14/28-day prediction and feature representation; a stage-specific rolling out-of-sample result reached about 65% directional accuracy, not a universal performance claim.
• Signal delivery: Built standardized model-management and API/JSON layers for predictions, factor exposures, and strategy weights.

🔬 Project Experience

Hong Kong Jockey Club Odds and Betting Behavior Statistical Modeling

Project Initiator | Oct 2025 – Dec 2025
CityU course project

• Collected and cleaned historical race, odds, and betting-type data, then applied statistical inference and learning to place and place-Q scenarios.
• This was a small-scale CityU course project; results should not be interpreted as stable betting ability or live trading returns.

Experimental Study on Shear Strength of Unsaturated Soil over a Wide Suction Range and Its Prediction

Core Team Member | National Natural Science Foundation of China related project 📄 Journal Paper (Rock and Soil Mechanics)

• Implemented high-accuracy fitting in MATLAB (above 98%) and rewrote core functions in C, supporting 200k+ data points with about 300% higher computational efficiency.
• Used Python / Origin for analysis and visualization, and contributed to two equipment-related patent outcomes.

Study on the Generation of Multi-rule Fractals

Outstanding Graduation Thesis

• Built a complex-dynamical-system-driven multi-rule fractal model and implemented real-time interactive rendering.
• Used Taichi / Python for GPU acceleration, achieving about 100× improvement over the original implementation.

Epidemiological Modeling of COVID-19

Mathematical modeling competition project | Mar 2022

• Built an SIERR-T model incorporating media and rumor dynamics, with sentiment and rumor analysis on Weibo comments.
• Analyzed policy effects using OxCGRT data.

📚 Publications & Patents

• [J.1] NIU Geng, ZHU Xiao-feng, LI Jun-xing, LÜ Meng-yuan, AN Li-qi, CHEN Zi-han. (2022). Experimental study on shear strength of unsaturated soil over a wide suction range and its prediction. Rock and Soil Mechanics, 44(12), 3349–3359. DOI: 10.16285/j.rsm.2022.2005. (IF=3.721, EI, CA, JST, CSCD, Peking University Core Journal)
• [P.1] NIU Geng, LI Jun-xing, ZHU Xiao-feng, TAN Yong-ming, MIAO Yu-song, ZHAO Shi-jun, WU Di. (2024). Triaxial Apparatus for Unsaturated Soil Based on Dialysis Method Suction Control. Chinese Patent, CN117368002A. Application No. CN202311298912.7, Application Date: 2023-10-09, Publication Date: 2024-01-09.
• [P.2] NIU Geng, KONG Liang, ZHU Xiao-feng, LI Jun-xing, AN Li-qi, LÜ Meng-yuan, CHEN Zi-han. (2022). Direct Shear Apparatus for Unsaturated Soil with Erosion Monitoring and Control System. Chinese Patent, publication CN115452613A; grant CN115452613B. Application No. CN202211220042.7, Application Date: 2022-10-08, Publication Date: 2022-12-09.

🏅 Honors & Awards

• 🥈 Chinese Mathematics Competitions (CMC), National Second Prize (Jan 2023)
• 🥈 CUMCM Mathematical Contest in Modeling, Shandong Second Prize (Nov 2022)
• 🥈 Shandong Provincial Mathematics Competition, Shandong Second Prize (Nov 2022)
• 🥉 MathorCup Mathematical Modeling Challenge, National Third Prize (May 2022)
• 🥈 Shandong Provincial Physics Competition, Shandong Second Prize (Nov 2021)

🛠 Skills

Toolchain Highlights

• Quantitative research: Pandas, NumPy, Polars, DuckDB, LightGBM, XGBoost, MAVE, VAE, time-series cross-validation, walk-forward validation, backtesting.
• Data infrastructure: Apache Parquet, S3-compatible object storage, MySQL, Redis, JSON/API delivery, partitioned and incremental data pipelines.
• Research engineering: PyTorch, Hugging Face Transformers, MMSegmentation, PaddleSeg, MONAI, TorchGeo, nnU-Net, TotalSegmentator.
• Systems and delivery: Linux/Unix, Git/GitHub, Slurm, Singularity, Docker, GitHub Actions, YAML configuration, CLI workflows, logging, and experiment tracking.
• AI-agent workflows: Claude Code, Codex, Cursor, GitHub Copilot, Dify, MCP, and business-facing automation design.

🌏 Languages

• Mandarin Chinese (Native)
• English (B2)

🎯 Interests

🏞 Hiking · 🏋️‍♂️ Fitness · 🏍 Motorcycle · 📷 Photography · 💻 Coding

---

Last Updated: Jul. 2026