Portfolio
Check out my featured projects
This deep learning project leverages Convolutional Neural Networks (CNNs) to classify MRI images as either tumor or non-tumor. Using a labeled dataset of brain scans, the model was trained with data augmentation and regularization to improve generalization. The final model achieved 97% validation accuracy and 98% test accuracy, demonstrating strong potential for clinical decision support.
Key challenges included limited data availability and class imbalance, which were addressed through aggressive augmentation (rotation, zoom, shift) and dropout layers to prevent overfitting. The model's performance was evaluated using metrics like accuracy, precision, recall, and confusion matrices.
Key Results:
Training Accuracy: 98.7%
Validation Accuracy: 97.0%
Test Accuracy: 98.2%
Precision (Tumor): 97%
Recall (Tumor): 96%
F1-Score: 96.5%
Financial-Analysis-Automation-with-LLMs is a Streamlit-based AI app that lets users analyze stocks using natural language queries. It combines real-time market data with large language models to deliver top stock recommendations, key financial metrics, AI-generated comparison summaries, and interactive performance visualizations — tailored to investor needs.
Key challenges included aligning structured financial data with natural language output, handling ambiguous user queries, and creating a responsive UI for multi-layered insights.
Codebase-RAG is an AI-powered tool that lets users interact with any public GitHub repository through a chat interface. It clones and indexes the repo using Pinecone's vector search, allowing users to ask natural language questions and receive detailed insights about the project's code structure, components, and functionality.
Key challenges included embedding large codebases efficiently, managing context-aware responses using LangChain, and building a smooth chat experience tailored to developers. This tool streamlines the process of understanding complex or unfamiliar codebases.
Pentagram is a full-stack web app where users can generate AI images from text prompts using Hugging Face diffusion models and share them with the creative community. The app features a responsive design, a gallery of generated posts, and a prompt-based creation flow that encourages creativity and interaction.
Challenges included integrating Hugging Face inference endpoints, managing image generation latency, and ensuring real-time updates to the shared feed using MongoDB and Cloudinary. The result is a polished, end-to-end creative AI platform.
VidStream is a responsive TikTok-style web app that allows users to upload, view, like, and comment on short-form videos. Built with Next.js and TypeScript, and powered by Sanity.io, it supports Google OAuth login, video uploads, real-time interactions, and category-based content discovery.
Challenges included building a smooth video upload flow with Sanity's asset API and managing user authentication and global state using Zustand. The app delivers a polished, mobile-friendly experience that closely mirrors TikTok's core features.
ChurnVision is a machine learning-powered web app that predicts customer churn by analyzing behavioral and demographic data. The system trains and compares multiple models—including XGBoost, Random Forest, and SVM—selecting the best-performing one (XGBoost, 85.4% accuracy) for deployment. The app provides real-time predictions, interpretable churn explanations, and generates personalized email retention strategies using the OpenAI API.
Key challenges included handling class imbalance, tuning hyperparameters for multiple classifiers, and integrating predictive insights with user-friendly output in a deployed Streamlit app.
WriteFlow AI is a writing assistant designed to help students generate and polish academic content. Users can input prompts and receive structured, high-quality writing suggestions powered by OpenAI. The app features a clean UI, customizable tone/style options, and supports exporting content for academic use.
A major challenge was designing prompts that adapt to diverse academic writing needs. This was addressed through modular prompt engineering and context-aware input structuring.
ProfAI is a web app that scrapes RateMyProfessors pages submitted by users and uses AI to generate smart summaries and recommendations. It integrates Pinecone and RAG (Retrieval-Augmented Generation) to deliver fast, relevant insights through semantic search.
Key challenges included handling inconsistent page structures during scraping and integrating Pinecone with LangChain for accurate, real-time data retrieval and analysis.
A full-stack SaaS web app that uses OpenAI to generate intelligent flashcards from user-provided text, helping learners retain information more efficiently. Features include authentication, credit-based usage limits, and responsive design for seamless use across devices.
A multilingual AI customer support chatbot with a modern UI, OpenAI-powered responses, and secure user roles.
About Me
Turning complex problems into simple design
5+
Years of experience
43+
Projects done
I am Bitania Yonas, a passionate software engineer with a keen interest in VR development, 3D environment design, and machine learning. My expertise spans across various programming languages and platforms, enabling me to tackle complex challenges and create immersive experiences.
With a strong foundation in computer science, I have successfully developed several projects, including educational VR environments, task management systems, and interactive AR interfaces. I thrive in collaborative settings, continuously learning and applying new technologies to deliver innovative solutions.
My education
Bachelor of Science in Computer Science
St. Thomas Aquinas College, Sparkill, NY
Major in Computer Science, focusing on software development, VR, and machine learning.
Work experience
Software Engineering Resident
Headstarter, New York, NY
- Building full stack AI projects in Python like churn prediction and brain tumor classification with neural networks and machine learning models and large language models such as Meta Llama 3.2 & Google Gemini
- Developed a full stack web application to perform multimodal RAG models over codebases using the GitHub API and Pinecone
- Deployed a stable diffusion image generation model using HuggingFace Transformers and Modal, integrated with Next.js frontend
- Collaborating with mentors for code reviews and weekly discussions, and mentored by a senior software engineer.
Software Engineer Fellow
Headstarter, New York, NY
- Built five full stack AI projects using React, TypeScript, and OpenAI API and deployed them in production environments.
- Developed a full stack web app using React, Next.js, TypeScript, and Clerk to assist users with writing documents with LLMs
- Developed collaborative workflows to enhance team-based project management and efficiency.
- Designed and presented AI solutions to senior engineers and key stakeholders for feedback.
Software Engineer
XLab Research, New York, NY
- Developed a virtual reality environment using Unity with C# scripting and 3D design.
- Implemented Cinemachine to enhance camera motion and user interaction.
- Built interactive elements to improve the educational experience within the VR environment.
- Designed the XLab website to showcase projects and research.
Software Developer
Tech Innovators, Detroit, MI
- Developed task management software using Python and Java to enhance team collaboration.
- Designed and organized coding workshops, leading team discussions on key development topics.
- Created and tested software to ensure system reliability and high-performance standards.
- Implemented process automation solutions to significantly increase overall productivity.
Software Engineering Tutor
YMCA, Addis Ababa, Ethiopia
- Created foundational web development skills for high school students by teaching HTML.
- Implemented coding workshops, providing hands-on guidance on programming concepts and techniques.
- Designed engaging STEM activities that enhanced critical thinking and problem-solving skills.
My favourite tools
Figma
vs code
Xcode
anaconda
Blender
HTML5
CSS3
Notion
Contact
Let's make something awesome together!
Done!
Thanks for your message. I'll get back as soon as possible.