Personal Resume
🔗 Online Bi-Directional Notes (Full Version)
Problem-Driven Full-Stack Solver | Skilled at Building Migratable AI & Data Application Systems
Keywords: Top-tier Chinese undergraduate, Master’s from overseas university, 10 years of experience in data analytics & algorithm applications, strong mathematics foundation, fluent in English, end-to-end project delivery
Ideal Roles: Data Mining, Machine Learning, AI Application Development
Target Industries: FinTech, Smart Manufacturing, Internet Platforms, Quantitative Investment
Personal Information
Name: Zhang Yuxiang | Education: Master's Degree |
Gender: Male | Date of Birth: 1988-12-22 |
Phone: 15982033650 | Email: yazyx88@163.com |
Education
-
2012.02—2014.11: Australian National University, Actuarial Science, Master’s Degree
-
2007.09—2011.06: Sichuan University, Applied Mathematics (Major), Finance (Second Major), Dual Bachelor’s Degrees
Work Experience
2024.07 - Present | AI Project Development & Frontier Learning
Focused on AI application development and technical exploration, I independently built several end-to-end systems, covering AI workflows, knowledge base RAG systems, intelligent customer service agents, and fine-tuning of generative (QA) and classification (intent recognition) models. I completed full-stack implementations—including backend logic, frontend interface, algorithm integration, and server deployment—to deliver modular, reusable, and production-ready AI applications.
These systems have been applied in multiple real-world scenarios, including:
-
A knowledge-based RAG system for handling user inquiries, deployed at Kan Village InfoHub, integrating embedding models and LLMs;
-
A legal document semantic retrieval project in collaboration with a law firm;
-
Two fine-tuning projects for the insurance industry: one for generative QA and one for multi-class intent recognition. Both projects covered the full pipeline from data construction to LoRA-based GPU training and local inference, significantly improving task-specific accuracy;
-
Additional launched projects include an AI-driven news summarization workflow, a portfolio management platform, and a Uplift-based precision marketing analysis system.
Through these hands-on projects, I’ve gained deep practical experience with large language models (LLMs), Retrieval-Augmented Generation (RAG), AI workflow design, and intelligent agents—continually exploring how AI can empower data understanding, knowledge management, and complex decision-making.
Keywords: Large Language Models, Fine-Tuning, RAG QA, Workflow, Multi-Platform Deployment, Reusability.
2020.03 – 2024.06 | Qianxing Private Fund | Strategy Researcher & Partner
-
Responsible for quantitative strategy research and live trading; led factor mining, strategy optimization, and backtesting framework development
-
Independently developed trading strategy based on genetic algorithms, achieving a 95% cumulative return over 14 months (vs. Shanghai Composite Index -11.26%), becoming the core profit driver contributing over 90% of revenue
-
Led internal AI tech adoption, exploring neuroevolution methods for high-dimensional feature modeling
-
As a partner, also engaged in investment decision-making, team management, and external collaboration
Keywords: Quantitative Investment, Strategy Development, Machine Learning, Python, Backtesting
2018.03 – 2020.01 | Ant Financial | Senior Data Analyst
-
Focused on data modeling and analysis for intelligent customer service and user operations, supporting millions of users
-
Led chatbot capability analysis, developed multi-dimensional trend decomposition algorithm, and filed Alibaba internal patent
-
Promoted development of intelligent audience targeting system using interpretable algorithms to enhance marketing precision
-
Independently designed and built end-to-end anomaly monitoring system to improve operational responsiveness
Keywords: Data Analysis, Trend Modeling, Association Mining, Anomaly Detection, Python, Machine Learning
2014.07 – 2017.10 | Kan Village Information Port & Qixin Education (Australia) | Operations & Data Analyst
-
Managed website operations and data analysis at Kan Village Information Port; independently built a WeChat chatbot
-
At Qixin Education, led analytics system setup and BI dashboard development
-
Initiated use of ML models (regression, decision trees) for solving real business problems
Keywords: Data Operations, BI Dashboarding, Intro to Machine Learning, Web Scraping, Data Visualization
Technical Skills
-
Over 10 years of experience in data mining and algorithm application, covering digital operations & data analysis, optimization algorithms & decision analysis, machine learning & data mining, deep learning, and AI-related applications.
-
Proficient in commonly used data analysis and mining methodologies, with strong abilities in problem identification and solution design; able to quickly propose efficient and practical solutions based on real-world needs.
-
Skilled in scientific methods such as randomized A/B testing and Uplift causal modeling; capable of independently executing precise user targeting, phased rollout, and effect evaluation for user acquisition and retention scenarios.
-
Expert in mainstream machine learning algorithms including logistic regression, decision trees, random forests, XGBoost, clustering analysis, DeepFM, and neural networks; well-versed in deep learning techniques and cutting-edge AI technologies.
-
Familiar with AI development involving large language models (LLMs), including AI workflows, Retrieval-Augmented Generation (RAG), and LoRA-based fine-tuning, realizing full-stack deployment from data to intelligent automation.
-
Knowledgeable in knowledge graph construction and applications, including entity extraction, relationship modeling, graph algorithms, graph databases, and visualization.
-
Understanding of PySpark and distributed data processing frameworks.
-
-
Able to independently complete full-cycle development of small-scale systems,
-
Including backend logic, frontend interface, algorithm integration, database design, and server deployment to deliver full product solutions.
-
Technical stack includes Python, MySQL, Linux, and basic server operations, applied to areas like productivity tools and information management systems.
-
-
Mathematics & English
-
Solid foundation in mathematics; capable of reading academic papers in English and implementing complex algorithms, with proven ability to rapidly apply them in real-world projects.
-
Over 5 years of study and work experience overseas; fluent in English and adept at cross-cultural and cross-regional communication and collaboration.
-
-
Strong learning agility and adaptability; excels at identifying the core essence of problems and designing scalable solutions—solving one problem often unlocks many.
Project Highlights (Selected)
📊 Multi-Dimensional Rate Decomposition System | Ant Financial | 2019
Background: Traditional one-dimensional analysis failed to explain root causes in conversion rate or satisfaction drops, especially with multi-dimensional business context.
Solution: Introduced LMDI (Logarithmic Mean Divisia Index) and developed automated Python scripts to decompose rate change into numerator/denominator contributions across dimensions.
Impact: Launched as a formal monitoring product for chatbot, satisfaction, and FCR metrics, covering daily millions of requests. Algorithm was filed as internal Alibaba patent.
This method solves any numerator/denominator metric decomposition across two time periods, ideal for interpreting conversion, escalation, and satisfaction rate changes. It helps pinpoint root causes with high explainability.
🔄 Follow-up (2025): A complete web product was built with frontend/backend/database integration.
→ Access Demo|→ View Report
👥 Interpretable Audience Segmentation | Ant Financial | 2019
Background: Traditional audience selection relied on manual heuristics or black-box models with low explainability and usability.
Solution: Adapted the Apriori algorithm to business needs, built a complete auto-pipeline for audience rule mining based on data preprocessing, feature transformation (WOE/IV), and rule output.
Impact: Applied in a major insurance campaign where >50% targeting rules were directly adopted. AB-test improves average 12%. Reused across retention, NPS, satisfaction feedback, etc.
This method applies to all binary classification problems (e.g., churn vs retain, convert vs not), focusing on interpretable rule discovery rather than black-box prediction.
🔄 Follow-up (2025): A web UI with parameter configuration and rule visualization was built.
→ Access Demo
📈 Complaint Trend Insight System | Ant Financial | 2019
Background: Complaint data volume was massive and unstructured, making it hard to detect meaningful trends across dimensions like product/channel/time.
Solution: Implemented “Top-K Insight” extraction algorithm combining linear regression, R², p-value scoring, and composite weighting to highlight sharp trend shifts.
Impact: Completed development during the final month at Ant Financial. The full pipeline for data analysis and insight extraction was validated end-to-end. The solution has strong reusability and can be flexibly applied to scenarios such as sales trend analysis, operational monitoring, and customer support volume fluctuation detection.
Suitable for any numeric metric (e.g., volume, complaint rate) needing multi-dimensional time trend tracking and prioritization. Helps uncover emerging or fading signals from complex noise.
🔄 Follow-up (2025): Integrated with LLMs for insight generation and automated summary reports.
→ View Report
🧬 Genetic Evolution Strategy Engine | Qianxing Private Fund | 2021–2022
Background: Traditional feature mining was manual and lacked generalizability. ML models were opaque and not suited for high-speed strategy deployment.
Solution: Built a multi-objective genetic algorithm framework optimizing return, drawdown, and complexity, with backtest engine + data generation + live scheduling.
Impact: Achieved 95% return over 14 months, becoming the company’s top-performing strategy (>90% revenue contribution).
Designed for any optimization task with explicit objectives but complex, non-differentiable search space—e.g., strategy selection, feature engineering, resource allocation.
🧠 Neuroevolution Experiment | Qianxing Private Fund | 2022
Background: Genetic algorithms struggle with high-dimensional features. Neural networks offer expression power but lack global search guidance.
Solution: Combined Transformer-based scoring model with evolutionary search using 8×A100 GPUs for high-dimensional strategy optimization.
Impact: Technical validation completed. Although not deployed due to business shift, methodology serves as foundation for future hybrid strategy generation.
Applicable to black-box optimization tasks where structure + nonlinearity matter, such as game AI, portfolio modeling, controller design.
📚 Embedding-based Knowledge Base Q&A | Kan Village | 2025
Background: Internal knowledge was scattered, hard to retrieve semantically, and lacked multi-format support.
Solution: Developed local system with embedding vectorization, FAISS search, and LLM-based RAG Q&A. Supported MD, PDF, Word, Excel, PPT files.
Impact: Applied to Kan Village customer support, personal notes, and law firm pilot.
Works for enterprise/internal document search, enabling semantic Q&A + multi-format ingestion with local deployment and privacy guarantees.
🤖 Intelligent Support Agent (LLM+Tools) | Personal | 2025
Background: Traditional support bots failed to flexibly handle diverse intents or trigger backend actions.
Solution: Built a LangChain-based agent combining intent recognition, FAQ retrieval, ticketing, complaint routing, and LLM-generated replies.
Impact: Completed prototype with modular interface, easily extendable to enterprise-grade customer service.
→ Access Demo
Applicable to multi-intent user scenarios requiring tool-based reasoning + response orchestration.
📰 LLM News Summarization & Aggregator | Personal | 2025
Background: Fragmented, emotional news content lacked structure and objectivity.
Solution: Built pipeline combining RSS/crawlers, LLM-based summarization, sentiment classification, and similarity clustering into a multi-stage summarization framework.
Impact: Deployed lightweight frontend showing curated daily topics with source/label/subscription.
→ View Site
Suitable for automated info pipelines with AI-powered summarization, tagging, and deduplication.
Fine-Tuned Multi-Class Intent Classification System (Insurance Domain) | Personal | 2025
Developed an insurance-specific intent classification model based on Qwen1.5-1.8B, covering 60+ business intents. Achieved lightweight customization via LoRA fine-tuning, completing a full pipeline from data construction to training and inference. The system supports Top-K multi-intent prediction and deployment, demonstrating high accuracy in real-world tests, and is well-suited for intelligent customer service applications.
🌟 Compact & Practical Projects (Selected)
Project Library Management System | 2025
A project portfolio management system featuring multi-dimensional tag filtering and fuzzy search. Supports dynamic filtering by tags, ratings, and year, greatly improving efficiency in organizing and sharing personal projects. Fully functional and easily extensible to personal portfolios, project dashboards, or content display platforms.
(Currently used for personal projects – View Site, photography collection – View Site)
Markdown Visualization Tool | 2025
A lightweight front-end rendering framework for Markdown documents with online preview and structured presentation. Supports Mermaid diagrams, LaTeX, and charts, suitable for technical documentation, data reports, and academic notes. Helps with rapid sharing and visual expression of structured content.
Now used in project outputs and automated reporting (e.g., View Sample Report).
Also usable as a standalone Markdown renderer – Try it here
Supervisor Selection Platform | University Project | 2022–2025
Developed a supervisor-student matching system using the Streamlit framework, serving over 50 graduate advisors and 200+ students. Supports two-way selection, CV management, and system roles for students, supervisors, and administrators. Students browse advisor profiles and submit applications; advisors review and confirm candidates; admins monitor the platform and export data.
The backend uses a MySQL database with basic login authentication to ensure data security. The system has been in continuous use for 4 years and successfully digitizes the supervisor matching process, improving transparency and communication efficiency.
(Built for a friend’s academic project; currently still in use and available for demo after this round of matching)
Other Highlights
-
Systematic thinking & learning: Skilled in distilling core issues, creating mindmaps, quickly analyzing, decomposing, and summarizing problems
-
Fast learning & knowledge management: Highly curious, can acquire, structure, and form solutions under tight timelines. Maintains over 5000+ structured personal notes enabling rapid knowledge synthesis and gap-filling
-
Logic & competition: Chess champion in Sichuan Province (national second-class athlete), top 0.1% on chess.com, national math competition prize winner
-
Hobbies: Magic (Alibaba annual event performer, Canberra Magic Club member), Photography (captured & edited company finals), Sports (Badminton)
-
Team & organization: Former president of Sichuan University volunteer club; led large-scale service events
-
Academic Exchange & Perspective Expansion: Maintain ongoing communication with algorithm experts from leading tech companies such as Ant Group and ByteDance, focusing on the practical application of machine learning and large AI models. Stay aligned with cutting-edge trends and continuously broaden technical horizons.