Guided Khoury graduate Computer Science students through their job search process, offering personalized advice and support, while also providing mentorship to aid in their academic advancement and comprehension of computer science principles

• Facilitated the creation of a company-wide public GoLang API by employing SWIG to seamlessly convert C++ code into GoLang• Established two separate data pipelines for processing user query logs using GoLang, and Protocol Buffers• Designed and implemented a monitoring dashboard to track query metrics, such as query latency, utilizing SQL• Defined metric baselines to create alerting indicators and set up a trigger attention system for monitoring unhealthy metrics

• Developed APIs supporting Agents for generative AI services, including the AI Front-End Code Generator, leveraging Langchain and LLMs to enhance client capabilities and improve business outcomes• Independently fine-tuned pre-trained models, including Alpaca, Falcon, Llama, WizardCoder, and others, to optimize their performance for code generation• Designed and developed a landing page and marketing email template generator with React and Node.js, implementing RESTful APIs using the best fine-tuned model

• Created an EC2 Instance tracker in GoLang, running in all AWS regions, and automated the homogeneity metrics checking process, which decreased internal tickets and cut down operational overhead• Added some IAM roles by building a code package written in TypeScript, and improved the speed of credentials generation • Designed a pipeline with LPT in Ruby, and deployed the project into this pipeline successfully

• Cooperated in developing the VUV (Vacuum Ultra-Violet) Spectra Library Viewer software, enabled researchers at the college and all staffs in the sponsored company to efficiently view the spectra of chemicals• Built a web scraper in Python to automatically collect hourly data on PubChem, Alpha Vantage, and Stocktwits, used as training set for machine learning models• Established various models including SVM, Random Forest, and MLP for stock classification, and enhanced the classification model accuracy from 72.1% to 83.7%• Accomplished an algorithm for feature selection, reduced the number of stock features from hundreds to 10, and deducted the cost of computation over thousands of dollars• Mined data on Stocktwits to extract text features with Scikit-learn and classified them into three categories – bearish, bullish, and neutral, which improved the understanding of stock markets for supervisors