Developed a machine learning pipeline to predict customer churn and identify factors associated with customer attrition. I performed data cleaning, exploratory analysis, feature preparation, model training, and evaluation across multiple classification algorithms. I compared Decision Trees, Random Forest, Gradient Boosting, XGBoost, and LightGBM models and evaluated their performance to identify a suitable predictive approach. The project strengthened my experience in model comparison, validation, feature engineering, and translating predictive modeling into actionable business insights.

Contributed to AfterQuery’s Project Pluto, developing high-quality benchmark tasks designed for training and evaluating large language models. I created realistic software engineering, DevOps, data engineering, and AI scenarios, wrote detailed task specifications and evaluation criteria, and developed reproducible environments and validation workflows. I executed tests locally, investigated edge cases and dependency issues, and ensured submissions were technically correct, reproducible, and compatible with automated evaluation pipelines. The work required structured reasoning, attention to detail, technical documentation, software debugging, and independent execution in a remote asynchronous environment.

Contributed to AfterQuery’s Project Silver by developing realistic software repositories designed for AI training and evaluation. I built production-style codebases with authentic Git histories, implemented features according to benchmark specifications, configured dependencies and testing workflows, and validated repositories against automated evaluation systems. I also diagnosed software defects, dependency problems, and configuration issues while maintaining clean, reproducible, and well-documented repositories. This project strengthened my ability to combine software engineering judgment with the requirements of AI training and evaluation.

Developed an LLM experimentation platform for interacting with language models through a web interface and backend API. I designed the application architecture using Next.js and FastAPI, integrated local LLM inference, and implemented API-based communication between the frontend and backend. The project provided practical experience in LLM application development, API design, asynchronous workflows, model integration, and building user-facing AI interfaces.

Built and contributed to digital and AI-powered solutions through Marixion, combining software engineering, AI, data, and modern web technologies to address practical business problems. My work has involved developing web applications, backend services, APIs, database integrations, and AI-enabled functionality. I have worked across the development lifecycle, including requirements analysis, architecture, implementation, debugging, testing, deployment, and technical documentation. Available online at marixion.com

MEDAITUTOR is an AI-powered learning platform designed to help students interact intelligently with their study materials. I developed the backend using Python and FastAPI and implemented a RAG pipeline that processes uploaded documents, generates embeddings, performs semantic retrieval, and provides context-aware responses. I also developed AI-powered features such as document-based chat and automated flashcard generation, integrating the application with PostgreSQL/Supabase and vector search. The project involved designing APIs, debugging backend services, integrating LLM capabilities, and building a practical AI application from concept to working system. It’s publicly available at medaitutor.com