Christina Hailu

AI-Enhanced Closed-Loop Communication Tool for Medical Training

Developed as part of my research internship at the University of Michigan through the AURA program, this project formed the foundation of the published study “AutoCLC: Towards Automated Assessment and Feedback on Closed-Loop Communication in Team-based Healthcare Simulation Training” (ACM Transactions on Computing for Healthcare, 2025).

The system AutoCLC leverages large language models to automatically detect, classify, and visualize Closed-Loop Communication (CLC) patterns within simulation-based emergency medicine training. It empowers instructors to assess communication efficiency and provide quantitative, example-based feedback to trainees in real time.

• I contributed by building an interactive web visualization dashboard for annotating and analyzing team interactions between medical trainees.
• Implemented data pipelines to process real-time communication transcripts from VR-based simulations and produce structured datasets for LLM analysis.
• Integrated GPT-4-powered NLP modules that automatically identify Call-Out and Check-Back statements with over 78.9% precision, achieving near-human annotation quality.
• Designed intuitive dashboards for instructors to visualize CLC performance, including rates of effective, incomplete, and missed communication loops.

This project bridges AI, human-computer interaction, and medical education demonstrating how explainable AI can enhance teamwork training and improve decision-making under pressure.

Image-to-Text Medical Test Analyzer

Contributed as a Backend Engineer for the Medi Map: Medical Test Locator system, a full-stack platform that bridges patients, laboratories, and healthcare providers by transforming medical reports into structured, queryable digital data.

The system combines OCR, intelligent data extraction, and geolocation to make diagnostic results searchable and actionable in real time. My primary focus was building the Node.js backend that powers all data ingestion, analysis, and service integration pipelines.

• Designed and deployed a modular backend architecture using Node.js and Sequelize ORM, ensuring scalability and maintainability across all services.
• Implemented advanced OCR pipelines with Tesseract.js, pdf-parse, and the OCR.Space API to extract patient, doctor, and test details from scanned PDFs and images.
• Built authentication and admin modules featuring OTP verification, password reset, and secure role-based access control for data privacy compliance.
• Integrated a Google Maps–based geolocation engine that lets users locate nearby diagnostic centers offering specific medical tests in real time.
• Containerized the full system with Docker and configured PostgreSQL for structured data persistence and indexing of test results.

This project reinforced my expertise in OCR optimization, backend data modeling, and scalable API design, demonstrating how intelligent automation can advance healthcare accessibility.

Shared Decision-Making Dashboard

Designed and developed an interactive data visualization and conversational AI dashboard to support clinicians and researchers in exploring Shared Decision-Making (SDM) survey data collected from patients, radiation oncologists, and urologic oncologists.

The goal was to make complex survey datasets easier to interpret, enabling medical teams to identify communication gaps, compare perspectives, and improve care coordination.

• Built a React.js frontend integrated with FastAPI for data querying and visualization.
• Developed dynamic Radar and Violin charts to display trends and distributions across different participant groups.
• Integrated a conversational AI agent powered by LangChain and Groq’s LLaMA3-70B model to answer natural language questions and generate on-demand analyses.
• Used ChromaDB and SentenceTransformers embeddings for semantic vector search, enabling intelligent data retrieval and summarization.

This project demonstrated how AI-driven data interaction can make medical insights more accessible and actionable, bridging the gap between data analytics and human understanding.

Decision Intelligence Interactive Dashboard

Developed an interactive data visualization dashboard that transforms complex datasets into actionable insights, empowering research and clinical teams to make informed, data-driven decisions with confidence.

The system enables users to dynamically explore data, uncover meaningful patterns, and visualize relationships through an intuitive, human-centered interface designed for clarity and adaptability.

• Engineered a modular React.js frontend integrated with a backend for real-time data exchange and analytics.
• Implemented flexible filtering and aggregation logic to support exploratory data analysis on large and complex datasets.
• Designed interactive chart panels that visualize trends, key performance indicators, and comparative results with precision.
• Applied user-centered design principles to ensure accessibility, usability, and a seamless visual experience across devices and contexts.

This project demonstrates the practical application of technology to build scalable systems that transform unstructured data into structured, queryable, and interoperable datasets for improved decision-making, accessibility, and operational efficiency.

Amharic Hate Speech Detection

Developed a deep learning model to detect hate speech in the Amharic language, addressing the growing challenge of online toxicity and the need for AI-driven moderation on Ethiopian social media platforms.

The system was trained on the publicly available Amharic Facebook Hate Speech Dataset containing over 30,000 labeled posts. A Convolutional Neural Network (CNN) architecture was implemented to identify contextual hate patterns and classify text as hate or non-hate with an accuracy of 66%.

• Performed extensive data preprocessing, including token filtering, word length normalization, and handling of morphologically rich Amharic script to prepare the dataset for deep learning.
• Designed and trained a CNN model using TensorFlow and Keras with embedding, convolution, max-pooling, dropout, and dense layers for binary text classification.
• Utilized 100-dimensional word embeddings to capture semantic relationships within the Amharic corpus vocabulary.
• Conducted model evaluation using precision, recall, F1-score, and confusion matrix analysis to assess detection reliability.
• Addressed challenges such as dataset inconsistencies, spacing errors, and morphological complexity that impacted model generalization.

This project highlights the importance of advancing low-resource NLP for African languages. By leveraging deep learning for Amharic, it contributes toward building safer and more inclusive online communication environments in Ethiopia.

AI-Powered Construction Equipment Rental Platform

Designed and engineered a full-stack construction equipment rental platform that connects contractors with suppliers through a secure, intelligent marketplace. The system enables equipment search, bidding, booking, and subscription management enhanced with AI-driven recommendations and real-time notifications.

As Team Lead and Backend Developer, I led a 5-member team and designed and developed the full backend architecture.

• Architected a modular, microservice-based backend with Node.js, PostgreSQL, Sequelize ORM, and Render Cloud for secure, scalable deployment.
• Implemented equipment management, real-time bidding, and booking workflows with integrated payment verification through Chapa API.
• Developed a recommendation engine that suggests relevant equipment using user history, rental frequency, and item ratings.
• Built a notification system using Firebase Cloud Messaging (FCM) to send booking and subscription updates in real-time.
• Integrated a phone-based OTP authentication system via Afromessage API for secure user verification and password recovery.
• Designed and exposed RESTful APIs for frontend (Flutter & web) integration, supporting user, renter, and admin roles.

This project integrates intelligent recommendations, secure authentication, automated bidding and booking workflows, and real-time notifications into a unified platform that streamlines equipment rental operations.

Grade Prediction System

Developed a machine learning model that predicts students’ academic performance to help educators and learners identify improvement areas and optimize study plans. The project uses real-world student data to uncover performance patterns and provide data-driven insights.

• Designed and trained regression models using Python and Scikit-learn to predict student grades based on attendance, test scores, and prior academic performance.
• Applied feature scaling and data preprocessing techniques to improve model accuracy and minimize bias.
• Conducted comparative evaluation between Linear Regression, Random Forest, and Gradient Boosting algorithms to select the most effective approach.
• Visualized results and prediction trends using Matplotlib and Seaborn for clearer interpretability.

The system demonstrates how predictive analytics can enhance learning strategies, helping both students and instructors make informed academic decisions.

Dimir Structure’s Website

Designed and developed a modern landing page website for a construction and architecture firm, focusing on clean design, intuitive navigation, and responsive layouts that enhance user experience across devices.

The project emphasized brand identity and visual storytelling to showcase the company’s services, completed projects, and design philosophy through a seamless, engaging interface.

• Created high-fidelity UI mockups and prototypes in Figma with a focus on minimalist, user-friendly design.
• Implemented responsive layouts and interactive components using HTML, CSS, and JavaScript.
• Applied modern design principles including consistent color schemes, spacing, and typography for brand cohesion.
• Optimized the site for fast loading and accessibility to ensure smooth client engagement.

This project strengthened my ability to translate UI/UX design concepts into functional, production-ready web experiences while maintaining visual elegance and performance.

Commercial Graduates Association (CGA)

Collaborated on the development of a backend system for the Commercial Graduates Association (CGA), a digital platform designed to manage member registrations, events, and communication efficiently. The project emphasized scalability, security, and performance.

My primary focus was implementing modular APIs, optimizing database queries, and ensuring smooth data flow between frontend and backend components to support high-volume user operations.

• Built RESTful APIs using Node.js and Express.js for membership management and event scheduling.
• Designed and optimized relational database models in PostgreSQL for efficient data access and reporting.
• Implemented authentication and role-based access control to ensure secure user interactions.

This project strengthened my expertise in backend architecture, API design, and database optimization, preparing me for scalable, production-grade systems development.

User Action Analytics Dashboard

Designed and implemented an interactive analytics dashboard to track and visualize user behavior over time, providing researchers and product teams with actionable insights into usage patterns and engagement trends.

The project focused on transforming raw event logs into meaningful visual reports, helping stakeholders monitor platform activity and identify key performance indicators (KPIs) through an intuitive, data-driven interface.

• Engineered a backend pipeline using FastAPI and PostgreSQL to aggregate and query user activity data efficiently.
• Developed dynamic dashboards with React.js and Chart.js to visualize metrics such as login frequency, feature usage, and engagement duration.
• Implemented filtering and time-based queries to allow flexible exploration of user actions by role, session, and time interval.
• Focused on data clarity and usability to ensure the insights were easily interpretable by both technical and non-technical audiences.

This project strengthened my coding skills while building a real-world solution.

GitHub Profile Finder

Built a Flutter application that enables users to search for GitHub profiles by username and view detailed information such as repositories, followers, and activity statistics fetched directly from the GitHub API.

The app demonstrates clean architecture, reactive state management, and seamless API integration using the BLoC pattern to handle asynchronous data flow efficiently.

• Developed using Flutter and flutter_bloc for robust state management and maintainable architecture.
• Integrated the GitHub REST API for fetching live user data, repositories, and activity details.
• Designed responsive and user-friendly UI components with real-time loading indicators and dynamic rendering of fetched data.
• Implemented clear error handling, search feedback, and repository data mapping to enhance usability and performance.

This project strengthened my skills in Flutter development, API integration, and state management while building a practical tool that connects developers with live GitHub insights.

Python Watch

Built a digital watch application entirely in Python using the Turtle graphics library, blending programming logic with creative visualization. The project showcases how simple animation and timekeeping can be combined to produce an elegant, real-time display.

Designed as an exploratory project to strengthen Python’s GUI and graphics capabilities, this interactive clock updates dynamically and demonstrates core programming concepts such as loops, time modules, and object manipulation in Turtle.

• Implemented the main watch face and hands using Turtle’s drawing functions with precise geometric positioning and rotation.
• Integrated the time module to synchronize real-world system time with smooth graphical updates.
• Focused on animation control and event-driven updates to ensure continuous and accurate rendering.
• Experimented with color customization and motion design to enhance visual appeal and user engagement.

This project reflects my curiosity and creativity in applying Python for visual problem-solving and interactive programming.