Built a large-scale, automated data pipeline project. Use intelligent crawlers to obtain a vast amount of documents from EDGAR and Municode. After being processed and analyzed by advanced NLP technology, the data is stored in a high-performance database to enable semantic search and comparative analysis across documents.

•  Crawled 4M+ lines of data (1TB)‌ from ‌EDGAR‌ and ‌Municode‌ reports using ‌Python‌ and ‌BeautifulSoup‌, implementing ‌virtual IPs‌ and ‌proxy rotation‌ to bypass rate-limiting and accelerate retrieval by ‌3x‌.

•  Built an ‌NLP preprocessing tool‌ with custom ‌tokenization‌ and ‌embedding generation‌ (Word2Vec/BERT), improving text processing efficiency by ‌40%‌.

•  Conducted ‌embedding-based comparative analysis‌ by integrating ‌EDGAR filings‌ with ‌CROPS datasets‌, enabling cross-document semantic search.

•  Designed an optimized ‌SQL database schema‌ for EDGAR data storage, leveraging ‌indexing‌ and partitioning to boost query performance by 30%‌.

A Bitcoin price prediction system integrating social media sentiment analysis and multi-model time series prediction has been constructed. Advanced machine learning and deep learning models are employed to enhance prediction accuracy. This provides reliable empirical analysis tools and model selection basis for quantitative trading strategies and market behavior research.

•  Utilized LSTM, BiLSTM, and DLinear for time-series analysis of Bitcoin price features. Improved prediction accuracy by 12% through hyperparameter tuning and optimizing model architectures.

•  Employed NLTK and VADER for sentiment analysis for over 100K of Twitter data, transforming text into structured data and removing 15% noisy data.

•  Designed and implemented automated scripts for data extraction, preprocessing, modeling, and visualization, Improved data analysis efficiency by 50%.

•  Used Matplotlib and Seaborn to generate heatmaps, boxplots, time-series plots for analyzing relation between time series, model choice and the prediction of Bitcoin price, improved analysis accuracy by 15%.

Developed a system using NLP and OpenAI’s GPT-3.5 Turbo API to analyze and extract insights from online product reviews, initially focusing on smartphones. Features include automatic summarization, sentiment analysis, and analytics to enhance consumer insights and inform product development strategies.

•  Coordinated the preprocessing of multiple product review datasets, cleaning and structuring the data to provide high-quality input for NLP tasks.

•  Integrated NLP and machine learning models, employing the Bag of Words model with stemming and lemmatization to process text reviews, enhancing text understanding accuracy by 20%.

•  Applied K-means clustering and the elbow method to group reviews by content similarity, utilizing OpenAI GPT-3.5-Turbo for detailed consumer sentiment analysis, enhancing the precision of insights on key product aspects.

•  Employed Ridge regression to enhance predictive accuracy by 15%, quantifying review content's impact on product ratings and analyzing consumer behavior, with detailed assessments of regression coefficients for product acceptance.

•  Leveraged OpenAI GPT-3.5-Turbo for sentiment analysis to enhance keyword ranking and model accuracy, improving the reflection of consumer opinions by 20%.

Developed a system to categorize and rate campus facilities, including academic buildings and dormitories. The system maintains detailed facility information and allows students to rate, enhancing campus engagement and feedback.

•  Developed a robust backend architecture with Spring Boot, enabling secure and efficient user interactions through RESTful APIs for operations like rating, bookmarking, and commenting.

•  Implemented a MongoDB schema to manage user, location, and rating data, utilizing GridFS for efficient binary data handling, boosting file storage and retrieval performance by 25%.

•  Configured AWS EC2 with load balancing to improve application stability and scalability, enhancing performance by 30%. Integrated CloudFlare CDN for reduced latency and increased protection against DDoS threats.

•  Developed a responsive React front-end with Axios for asynchronous API calls, enabling real-time interactions like voting and commenting without page reloads.

Developed multiple artificial intelligence agents to tackle complex problems in the Pac-Man game, optimizing for speed and strategy across various levels with multiple objectives.

•  Used pathfinding algorithms, DFS, BFS, Uni Uniform Cost, A* Search achieved 95% winning rate for Pac-Man, reduced 50% of source utilization.

•  Designed multi-agent search strategies, allowing effective coordination between Pac-Man and ghost agents, achieving a 30% increase in gameplay efficiency and 40% reduction in agent conflicts during simulations.

•  Applied probabilistic reasoning techniques to enhance AI decision-making, achieving a 30% increase in prediction accuracy for ghost movements in dynamic environments.

•  Implemented reinforcement learning techniques that increased AI agents' gameplay scores by 25%, optimizing strategies based on continuous feedback loops.

I built an open platform API from scratch, based on React, SpringBoot, Gateway, and RPC technology. Administrators can access and publish interfaces, as well as visualize the calling status of each interface. Users can enable interface calling permissions, browse interfaces, and debug online, and easily call interfaces through the client SDK.

•  Used Ant Design Pro to quickly build the basic page, slimmed down the original template and abstracted it into a reusable public template, increased the development efficiency by 80%.

•  Developed starters, the user only needs to worry about which SDK to use and which parameters to transfer, just as easily as calling the code they wrote themselves.

•  Used AOP, don’t modify the original interface code to implement function usage statistics, code decoupling, and unified processing.

•  Implemented RPC interface, used remote methods just like calling local methods, made it more transparent to users and reducing communication costs, supported HTTP and TCP/IP, improved performance by 20%.

•  Created and performed 50+ automated unit tests with Junit, expanding code coverage to 80%, minimized bugs introduced in new features deployment.