Chadlia Elamri

At Choose, I built AI-powered automation tools that reduced manual processing errors by 70%. I replaced existing Google Sheets workflows with a custom web interface and integrated APIs, improving efficiency. I handled daily scraping requests, developing resilient pipelines with rotating proxies and anti-bot measures to ensure reliable data extraction. Additionally, I deployed microservices on GCP Cloud Run with automated CI/CD pipelines and delivered reusable internal packages following Domain-Driven Design principles, improving maintainability and scalability of internal systems.

- Improved the performance of Lavague, a web automation agent, through prompt engineering and testing various language models (LLMs) like GPT-4 and LLaMA. - Enhanced Lavague’s task automation by adding functions to the pipeline and optimizing its interaction with LLMs. - Contributed to the development of a web application that transforms RSS feed articles into videos using AI, with technologies like Python, Flask, React, and Remotion. - Implemented web scraping with BeautifulSoup and Selenium to process content for video creation. Focused on optimizing LLM-based workflows to improve task automation and content generation.

During this summer internship, I had the incredible opportunity to spearhead the development of a food swapping real-time web application. This innovative platform empowered users to initiate sessions, invite friends, and cast their votes for their preferred meals. At the end of each session, the application would unveil the top three meal choices, making meal planning and social dining a breeze. It was an exciting and rewarding experience that honed my skills and allowed me to contribute to a dynamic project with real-world impact.

- Designed and implemented a machine learning system to identify anomalies in fences using surveillance footage, leveraging Convolutional Neural Networks (CNNs). - Employed transfer learning with pre-trained models such as Xception, VGG16, and ResNet50, fine-tuning them for the specific task of anomaly detection. - Optimized model performance by tuning hyperparameters like learning rate and batch size, achieving a 99% accuracy using ResNet50. - Used evaluation metrics like accuracy, precision, recall, and F1-score to compare model performance, utilizing confusion matrices for detailed assessments.